CrowdStrike has admitted to pushing out a bad software update, causing many Windows machines running the affected software to crash. The problem, apparently affecting its Falcon platform, brought down servers at airlines, locked up computers at banks, and hurt healthcare services.

“CrowdStrike is actively working with customers impacted by a defect found in a single content update for Windows hosts,” the company said Friday in a post to its blog titled “Statement on Windows Sensor Update.”

Mac and Linux versions of the software are unaffected, and the incident was not the result of a cyberattack, it said.

Apple has confirmed recent claims that it might have used subtitle data from YouTube videos to train one of its artificial intelligence (AI) tools, but says the tool is not used in Apple Intelligence.

Apple confirmed a report from Proofnews that it had used this YouTube data to train one of its models. The company explained that it did so to train the open-source OpenELM models released earlier this year. The information was included within a larger collection maintained by the EleutherAI non-profit company that supports AI research.

Apple used YouTube once, but not now

However, Apple told 9to5Mac that models trained using that information don’t power any of its own AI or machine learning tools, including Apple Intelligence. This was a research project originally created by Apple’s AI teams and then shared, including via the company’s own Machine Learning Research site.

What’s important is that it shows the extent to which Apple wants to be seen as keeping its promise that Apple Intelligence models are trained on licensed data.

But that’s not the big picture. As mentioned earlier in the week, Apple Intelligence does also train its models using “publicly available data collected by our web-crawler.”That admission reflects the extent to which tech companies are using information published online to create new AI products from which they subsequently profit.

Making public data private

The issue is that by turning other people’s creative works into data, and then profiting from that data, tech firms aren’t playing fair. 

Speaking to Proofnews, Dave Farina, the host of “Professor Dave Explains,” put it this way: “If you’re profiting off of work that I’ve done [to build a product] that will put me out of work or people like me out of work, then there needs to be a conversation on the table about compensation or some kind of regulation.”

To some extent, the focus on YouTube data distracts from that critical argument, which is that the generative AI (genAI) tools coming into common use today are likely to have been trained by information created by humans and shared online. That’s the kind of information picked up by webcrawlers, including Apple’s.

But data quality is a real issue here, and the search for the best data inherently means that the best data sources are the highest octane of fuels to power training AI.

The drive for quality means content is king

Consider just two of the challenges AI researchers face.

• Automated data grading systems might reject old, out-of-date, or false information, but some still gets through, which is why AI systems so often develop hallucinations (the current descriptor for fake information) or exhibit questionable morality (racist or gender-biased language).
• Data also has a finite lifespan. Facts can and do change over time, and maintaining high-quality data is an essential bulwark against the classic “garbage out” generated by irrelevant information, or high-grade information that becomes irrelevant over time.

What this means is that in their quest for high-quality information, AI companies inevitably seek high-quality data sources. When you translate that into activity picked up from the open public web, that in itself implies that creatives currently battling against tech firms for compensation for use of their material in training AI systems have a good point.

Because the best and most current information they create is worth something, both to the creators, those who consume it, and also to the people who own and train the machines that harvest their data from it. Indeed, given that AI by its nature becomes a tool directly available to everyone and across every supported language, it seems plausible to think the value of that information might actually grow once it is used to train an AI model.

So, while Apple might not be using YouTube data for its Apple Intelligence models, it will be using other data curated across the public web. And while Apple might at least try to avoid using data it should not exploit this way — and is honest enough to have responded to the current YouTube controversy — not every AI firm does the same. And once the machine is trained it cannot be untrained.  

Please follow me on Mastodon, or join me in the AppleHolic’s bar & grill and Apple Discussions groups on MeWe.

When generative AI popularity and marketing hype went into overdrive last year, just about every enterprise launched a wide range of genAI projects. And for various reasons, very few of them delivered the kind of return on investment that CEOs and board members had expected.

That meant that 2024 has become the year of AI postmortems and recriminations about why projects went sour and who was to blame. What can IT leaders do now to make sure that genAI projects launched later this year and throughout 2025 fare better? Experts are suggesting a radical rethinking of how ROI should be measured in genAI deployments, as well as the kinds of projects where generative AI belongs at all.

“We have an AI ROI paradox in our sector, and we have to overcome it,” said Atefeh “Atti” Riazi, CIO for media enterprise Hearst, which reported $12 billion in revenue last year. “Although we have [years of experience] measuring the ROI for IT on lots of other projects, AI is so disruptive that we don’t really yet understand its impacts. We don’t understand the implications of it long term.”

When boards push down genAI mandates — and LOBs go rogue

After OpenAI captured the attention of the industry when consumer fascination with ChatGPT surged in early 2023, Conor Twomey observed a “wave of euphoria and fear that swept over every boardroom.” AI vendors tried to take advantage of this euphoria by marketing their own version of FUD (fear, uncertainty, and doubt), said Twomey, head of AI strategy at data management firm KX.

“Every organization went down the same path and said, ‘We don’t know what this thing is capable of.’”

That sparked a flood of genAI deployments ordered from boards of directors and, to a lesser extent, CEOs. This was happening to an extent that has not been seen since the early days of web euphoria around 1994.

“That was something different with generative AI, where a lot of the motion came top-down,” said Rajiv Shah, who manages AI strategy for Snowflake, a cloud data storage and analytics service provider. “Deep learning, for example, was certainly hyped up, but it didn’t have the same top-down pushing.”

Shah says this top-down approach colored and often complicated the traditional requirements for ROI analysis prior to major rollouts. Little wonder that those rollouts failed to meet expectations.

And mandates from above weren’t the only source of pressure IT leaders faced to push through genAI projects. Many business units also brought AI ideas to IT, and IT pointed out why they would be unlikely to be successful. And those departments often said, “Thanks for the input. We are doing it anyway.”

Such projects tend to shift focus away from companies’ true priorities, notes Kelwin Fernandes, CEO at AI consultant NILG.AI.

“I see genAI being applied in non-core processes that won’t directly affect the core business, such as chatbots or support agents. These projects lack support and long-term engagement from the organization,” Fernandes said. “I see genAI not bringing the promised ROI because people moved their priorities from making better decisions to building conversational interfaces or chatbots.”

Inflated expectations, underestimated costs

Early genAI apps often delivered breathtaking results in small pilots, setting expectations that didn’t carry over to larger deployments. “One of the primary culprits of the cost versus value conundrum is lack of scalability,” said KX’s Twomey.

He points to an increasing number of startup companies using open-source genAI technology that is “sufficient for introductory deployments, meaning they work nicely with a couple hundred unstructured documents. Once enterprises feel comfortable with this technology and begin to scale it up to hundreds of thousands of documents, the open-source system bloats and spikes running costs,” he said.

“Same goes for usage,” he added. “When genAI is inserted into a workflow ideal for a subset of users and then exponentially more users are added, it doesn’t work as hoped.”

Patrick Byrnes, formerly senior consultant for AI at Deloitte and now an AI consultant for DataArt, attributes some of the inflated ROI expectations for generative AI projects to the impressive performance delivered by the earliest genAI applications.

“If you go into Gemini or ChatGPT and ask it something basic, you can get an incredible response right away,” he said. Expecting similar results on a larger scale, “some enterprises did not start small. Right out of the gate, they went with high-impact customer facing efforts.”

Indeed, many of the ROI shortcomings with genAI deployments are a result of executives not thinking through the rollout implications sufficiently, according to an executive in the AI field who asked that her name and affiliation not be used.

“Automation driven by AI leads to productivity gains, but often the cost to enable it is overlooked,” she said. “Enterprises focus on model development, training, and system infrastructure but don’t accurately account for cost of data prep. They spin up massive data sets for AI, but small errors can make it useless, which also leads employees to mistrust outputs, leading to costs without ROI.”

Another overlooked factor, she noted, is that many AI vendors are currently focused on customer acquisition, keeping costs down in the short term. “Then they will ratchet up prices with an eye toward profitability, which will lead to higher costs for enterprise users in the future.”

Those costs are not likely to get meaningfully better by 2025. IDC noted that the costs with generative AI efforts are extensive.

“Generative AI requires enormous levels of compute power. NVIDIA’s workhorse chip that powers the GPUs for datacenters and the AI industry costs ~$10,000 per chip,” the analyst firm said in a September 2023 report. “Operational costs are in the range of $4 million to $5 million monthly, and businesses expect model training costs to exceed $5 million. Added to this are electricity costs and datacenter management.”

The hallucination challenge

On top of all this is the fact that genAI periodically hallucinates, meaning that the system makes things up. That will deliver a bitter surprise if the company is trusting it to analyze critical data in healthcare, finance, or aerospace — and even if it is simply relying on genAI to accurately summarize what happened during a meeting.

For business managers who are used to trusting the numbers generated by a spreadsheet projecting revenue growth, that can be unsettling. Those executives are used to the projections failing because an employee’s assumptions turned out to be too optimistic, but they are not used to Excel lying about the mathematical result of 800 numbers being multiplied.

And it cuts into ROI because all generative AI output must be closely fact-checked by a human, erasing many of the perceived productivity gains.

Hearst’s Riazzi sees the genAI hallucination issue as temporary. “Hallucinations do not bother me. Eventually, it will address itself,” she said.

More importantly, she argues that business simply needs to apply the same supervision and oversight to genAI that it has for decades with its human employees, stressing that “people hallucinate as well” and coders have been known to write “buggy code.”

“Human error is already a big issue in medicine and patient care,” Riazzi said. “There is a lot of bad data out there, but there is no difference [in managing hallucinations] from what we are already doing today. We see a lot of data cleansing going on.”

NILG.AI’s Fernandes is doubtful that genAI hallucinations will ever go away, but he says that shouldn’t necessarily be a dealbreaker for any application. It is simply a matter of enterprises adjusting their thinking to deal with an imperfect reality, something they already have experience doing.

“We have quality assurance to reduce production errors, but errors still exist, and that’s why we have return policies and warranties. We use the QA process as a fallback plan of the factory errors and the warranty as a fallback plan of the QA,” he said. “All those actions reduce the probability of failure to a certain point. They can still exist; we have learned to do business with those errors. We need to understand — on each application — what the right fallback action is for an AI error.”

Looking for ROI in all the wrong places

Even when genAI succeeds, its results are sometimes less valuable than anticipated. For example, generative AI is a very effective tool for creating information that is generally handled by lower-level staffers or contractors, where it is simply tweaking existing material for use in social media or e-commerce product descriptions. It still needs to be verified by humans, but it has the potential for cutting costs in creating low-level content.

But because it often is low level, some have questioned whether that is really going to deliver any meaningful financial advantages.

“Even before AI, the market for mediocre written and visual content was already fairly saturated, so it’s no surprise that some enterprises have discovered there is limited ROI in similar mediocre content generated by AI,” said Brian Levine, a managing director at consultant Ernst & Young.

What ROI should look like for enterprise genAI

KX’s Twomey questioned whether many senior enterprise executives have a realistic handle on what ROI should mean in a generative AI rollout, especially in the first year where it is mostly an experiment rather than a traditional deployment.

“Enterprise deployment of genAI has slowed down — and will continue to do so — as enterprises experience an increase in costs that exceeds the value they are getting,” Twomey said. “When this happens, it tells me that enterprises aren’t understanding the ROI and they’re not appropriately controlling TCO.”

And therein lies the conundrum: How can executives appropriately control the total cost of ownership and appropriately interpret the return on investment if they have no idea what either should look like in a generative AI reality?

This gets even more difficult when secondary ROI factors are considered, such as market and customer/prospect perceptions, Twomey points out.

“This complexity with transiting — and scaling — AI workflows in production has been prohibitive for many enterprise deployments,” he said. “The repercussions are clear losses in time, money, and effort that can also result in competitive disadvantages, reputational damage, and stalled future innovation initiatives.”

It may even be premature to measure ROI monetarily for genAI. “The value for enterprises today is to practice, to experiment,” said DataArt’s Byrnes. “That is one of the things that people don’t really appreciate. There is a strong learning component to all of this.”

Focusing genAI

But while experimentation is important, it should be done intelligently. EY’s Levine notes that some companies are inclined to trust generative AI too much when it comes to methodology, allowing the software to figure out how to obtain the desired information. 

Consider the example of a large and growing retail chain that turned to genAI to figure out the best locations for its next 50 stores. Given insufficient guidelines, the AI went off the rails and returned completely unusable results, according to inside sources.

Instead of simply telling the AI to make recommendations for the best places to launch stores, Levine suggests that the retailer would be better served by coding very extensive and very specific lists of how it currently evaluates new locations. That way, the software can follow those instructions, and the chances of it making errors is somewhat reduced.

Would an enterprise ever tell a new employee, “Figure out where our next 50 stores should be. Bye!”? Unlikely. The business would spend days training that employee on what to look for and where to look, and the employee would be shown lots of examples of how it had been done before. If a manager wouldn’t expect a new employee to figure out how to answer the question without extensive training, why would that manager expect genAI to fare any better?

Given that ROI simply means value delivered minus cost, the best way to improve value is to increase the accuracy and usability of the answers provided. Sometimes, that means not giving genAI broad requests and seeing what it chooses to do. That might work in machine learning, but genAI is a different animal.

To be fair, there absolutely are situations where it makes sense to set genAI loose and see where it chooses to go. But for the overwhelming majority of situations, IT will see far better results if it takes the time to train genAI appropriately.

Reining in genAI projects

Now that the initial hype over genAI has died down, it’s important for IT leaders to protect their organizations by focusing on deployments that will bring true value to the company, say AI strategists.

One suggestion for trying to better control generative AI efforts is for enterprises to create AI committees consisting of specialists in various AI disciplines, Snowflake’s Shah said. That way, every single generative AI proposal originating anywhere in the enterprise would have to be run by this committee, who could veto or approve any idea.

“With security and legal, there are so many things that can go wrong with a generative AI effort. This would make executives go in front of the committee and explain exactly what they wanted to do and why,” he said.

Shah sees these AI approval committees as short-term placeholders. “As we mature our understanding, the need for those committees will go away,” he said.

Another suggestion comes from NILG.AI’s Fernandes. Instead of flashy, large-scale genAI projects, enterprises should focus on smaller, more controllable objectives such as “analyzing a vehicle’s damage report and estimating costs, or auditing a sales call and identifying if the person follows the script, or recommending products in e-commerce based on the content/description of those products instead of just the interactions/clicks.”

And instead of implicitly trusting genAI models, “we shouldn’t use LLMs on any critical task without a fallback option. We shouldn’t use them as a source of truth for our decision-making but as an educated guess, just like you would deal with another person’s opinion.”

More by Evan Schuman:

• Renegade business units trying out genAI will destroy the enterprise before they help
• AI managing AI that is monitoring AI: What could possibly go wrong?
• US makes new move to rein in China’s advanced chip manufacturing
• GenAI might be the least-trustworthy software that exists. Yet IT is expected to trust it.
• Privacy policies have gone insane. Doubt it? Consider Instacart

Anthropic released an iOS version of its popular chatbot Claude for the iPhone in May; now, it’s time for an Android version.

The artificial intelligence (AI) company announced the Android iteration on Tuesday.

With the help of Claude, users can have conversations in a number of languages, including English, German, French, Spanish and Italian. The new app reportedly can be used with all subscriptions, which includes Pro and Team. Business users have the option to sign up for a monthly subscription that costs roughly $31 per user. The minimum number of users that can be registered is five.

Users who can’t download the app from Google Play or the App Store, can still access Claude via the web at claude.ai .

On June 18, 2024, the first round of Copilot+ PCs arrived, including offerings from Microsoft, HP, Asus Acer, Dell, and Lenovo. I was lucky enough to land a Lenovo Yoga Slim 7x on June 21^st and have been digging into its capabilities and limitations ever since.

Over the weekend, I stumbled upon a situation that is both unsurprising and disturbing — namely, that there are very, very few image backup and restore tools that work with the ARM64-based version of Windows 11 24H2 that ships on all currently available Copilot+ PCs.

Searching for working image backup packages

Indeed, a concerted series of Google and Bing searches have turned up exactly three software programs that can back up and restore ARM64 versions of Windows (all of which run only on Snapdragon X CPUs at present, though AMD64 versions on Intel and AMD CPUs are expected in the next month or two).

Two of those three options are at least mildly questionable, as I’ll explain:

1. Zinstall FullBack is a full-featured backup and restore package that performs constant incremental backups to a local or networked drive, or into the cloud. Zinstall has been active in the ARM side of Windows backup since Microsoft released early versions of Windows on ARM (WoA) for the Surface Pro X in November 2019. The vendor offers a free 30-day trial, and then charges US$14.90 per month thereafter to use the software.

A bare-metal restore to a non-booting PC will first require a clean Windows install on that machine (I’d recommend an ARM64 ISO from UUP dump), and then installing the Zinstall application. After that, you can restore a backup from your collection of prior snapshots and overwrite the temporary install with that install to pick up where it left off.

2. Microsoft’s Backup and Restore (Windows 7) Control Panel item is still available in Windows 11 24H2. As you can see in this Microsoft Learn article, Windows 7 Backup and Restore has been deprecated since the release of Windows 8 in 2012. This tool is intended to restore existing Windows 7 backups to newer Windows PCs, but it can back up and restore newer versions as well. It’s not a production-grade tool.

3. Version 6.0 of the Veeam Agent (which works with the company’s various backup and replication enterprise-grade solutions) has been force-fit to back up on ARM-based CPUs as of March 2023 (see the end of this R&D Forums note). It can be restored using a Veeam Agent running on an X64 PC. Here again, this appears to be something of a kludge.

Just for grins, I checked all of the backup packages mentioned in Tim Fisher’s November 2023 Lifewire article 32 Best Free Backup Software Tools. None of them supports ARM64 CPUs, either.

The Windows Backup option

When I asked Microsoft to comment on the situation, a spokesperson pointed me to the Microsoft support page for the Windows Backup app built into Windows 11, indicating that this tool provides a backup and recovery solution for ARM-based PCs. It does, but not completely.

As I discuss in a recent article on the new backup, recovery, and repair tools in Windows 11, the Windows Backup app is undoubtedly a useful tool for backing up files and folders, apps, settings, and credentials and restoring same. But its restore operation is not as seamless as when using dedicated image backup software, and it doesn’t easily scale up for enterprise use. Indeed, it requires one-at-a-time reinstall of all Windows apps and applications (through links in the Start menu) to fully restore a Windows 11 PC to match its backed-up installation state.

In other words, making complete image backups that can be quickly and easily restored requires third-party image backup software.

Get real about backup and restore in Copilot+ PCs

Realistically, Zinstall FullBack appears to be the only viable option for backing up and restoring Copilot+ PCs with Snapdragon X Elite and Snapdragon X Plus CPU models. (Elite models include X1E-00-1DE, X1E-84-100, X1E-80-100, and the X1E-78-100 found in the Lenovo Yoga Slim 7X; Plus models include X1P-64-100.)

Buyers considering an investment in the current crop of Copilot+ PCs should ponder this potential limitation (among others) carefully. They should also consider that the upcoming collection of Intel- and AMD-based Copilot+ PCs will work with all currently available Windows 11-compatible image backup and restore tools and platforms.

The UK’s Competition and Markets Authority (DMA) on Tuesday announced the start of an inquiry into Microsoft’s hiring of employees from Inflection for its consumer AI group, and its approximately $650 million payment to license the company’s technology.

Inflection cofounders Mustafa Suleyman and Karén Simonyan, along with a number of other colleagues, joined Microsoft in March; Suleyman became executive vice president and CEO of the newly-formed Microsoft AI, Simonyan joined as chief scientist for the group.

At issue is whether Microsoft’s actions constituted what the CMA called “a relevant merger situation,” and if so, whether the moves will substantially lessen competition in the UK.

The inquiry begins July 17, with a decision on whether to proceed with further investigation to be announced Sept. 11.

“Regulators are right to challenge these practices, as they can potentially stifle innovation and competition,” said Phil Brunkard, executive counselor at Info-Tech Research Group, UK. “Start-ups thrive on the creativity and vision of their founders and the unique ideas they bring to the table.

“Competition is essential for fostering creativity and innovation, and we should encourage start-ups to grow independently, allowing competition to develop naturally,” he said. “This ensures that investment is based on merit, rather than allowing dominant players to define the rules of competition on their own terms.”

UK regulators aren’t the only ones to call the activities into question. In June, the US Federal Trade Commission (FTC) launched an investigation to determine whether there actually was an undisclosed acquisition through the hiring of key personnel and the licensing agreement.

The European Union is also closely monitoring developments. Reuters reported in April that EU antitrust chief Margrethe Vestager was watching to see whether other companies emulate Microsoft’s strategy of a talent and technology transfer in place of a formal merger. If it becomes a trend that circumvents the rules around mergers and competition, she told Reuters, she might act.

Since the release of Apple’s iOS 18 developer beta 2, Rich Communication Suite (RCS) support has come to messaging on iPhones. That means you can look forward to a more platform-agnostic messaging experience than before, making messaging between work colleagues, partners, and friends better than before — sometimes by satellite.

What is Rich Communication Suite (RCS)?

The Global System for Mobile Communications Association (GSMA)-defined RCS standard aims to improve on standard SMS messaging with the addition of a suite of features you usually find on platforms like iMessage or WhatsApp. That means support for group chat, file transfers, typing notifications and more.

Initial work by the GSMA identified some successful customer engagement, marketing, and event communications usage scenarios for enterprise users. While Apple was highly resistant to implementing the standard on its devices, it has now changed its mind, partly as regulators began to question the decision not to offer such support. 

Apple has made one recent reference to RCS. “When messaging contacts who do not have an Apple device, the iMessages app now supports RCS for richer media and more reliable group messaging compared to SMS and MMS,” the company said in June.

What does RCS support on iPhone?

At present, RCS promises support for higher quality photos and videos, audio messages, and larger file sizes for attachments. It also provides read receipts and typing indicators, cross platform emoji reactions, and location sharing. Users can expect:

• Group chat.
• File transfers.
• Typing notifications.
• Higher resolution photos and video. 
• Audio messages.
• Read receipts.
• Typing indicators.
• Location sharing.
• Cross platform emoji.

You will know when you’re in an RCS chat with an Android user because you’ll see a small grey label that says RCS Message in the text field.

Is RCS safe to use?

RCS is not as secure as iMessage but does provide better encryption than you’ll get using SMS. It is possible that Apple will implement a more secure version of RCS in time, but as things stand, the most secure messaging option remains iMessage because it delivers end-to-end encryption.

What this means for iPhone users

The first thoughts on how RCS works between iPhones and Android devices are pretty positive. The images you share will be high-res rather than deeply compressed. Read receipts and typing indicators flow between both platforms. Standard Tapback responses also work, meaning you can send reactions to messages using that system. 

You won’t get access to text formatting or some of the other new iMessage features — and RCS messages remain encased in green bubbles with an accompanying label that tells you this was a Text Message in the RCS format.

Apple’s hierarchy of texts

There is a hierarchy to how messaging is handled. That means if two Apple devices are used to communicate, they will use Apple’s iMessage, which continues to be the best messaging experience on iPhones. 

If an Apple device is communicating with an Android device, the exchange will take place over RCS, and if the carrier doesn’t support RCS or there is no active data connection the messaging all takes place over SMS. At the risk of sounding obvious, SMS lacks the more advanced messaging features you will find in either of the other standards, and Apple’s approach still means iMessage is the best option.

What is the road map for RCS improvement?

The RCS experience will improve over time. The GSMA Association last month finalized the latest update to the standard, adding support for replies and reactions and the ability to edit, recall, and delete messages sent earlier for both parties. 

The update also includes a tool to report spam messages and additional support for Custom Reactions, which may mean that Genmoji and Photomoji will become more cross platform. Apple is working with Google and members of the GSMA to improve the standard worldwide, which implies features such as the ability to edit and delete messages should be available via RCS at some point.

How do I enable RCS on my iPhone?

If you are running the latest iOS 18 beta you can enable RCS in Settings>Apps>Messages, where you should find an RCS toggle. If you don’t see that, it’s likely your carrier doesn’t yet support RCS on iPhones. To support the feature, carriers need to update some of their own settings, which are usually bundled within iOS updates. It is likely more carriers will introduce support for this by the time the iOS 18 ships.

Where is RCS available?

Apple only enabled RCS support on iPhones in the second iOS 18 beta and only on some US networks. That support has now been extended to other nations and some networks, including those in Canada, Spain, France, Germany.

Please follow me on Mastodon, or join me in the AppleHolic’s bar & grill and Apple Discussions groups on MeWe.

Apple Intelligence isn’t entirely Apple’s intelligence; just like so many other artificial intelligence (AI) tools, it also leans into all the human experience shared on the internet because all that data informs the AI models the company builds.

That said, the company explained where it gets the information it uses when it announced Apple Intelligence last month: “We train our foundation models on licensed data, including data selected to enhance specific features, as well as publicly available data collected by our web-crawler, AppleBot,” Apple explained.

Your internet, their product

Apple isn’t alone in doing this. In using the public internet this way, it is following the same approach as others in the business. The problem: that approach is already generating arguments between copyright holders and AI firms, as both sides grapple with questions around copyright, fair use, and the extent to which data shared online is commodified to pour even more cash into the pockets of Big Tech firms. 

Getty Images last year sued Stability AI for training its AI using 12 million images from its collection without permission. Individual creatives have also taken a stance against these practices. The concern is the extent to which AI firms are unfairly profiting from the work humans do, without consent, credit, or compensation.

In a small attempt to mitigate such accusations, Apple has told web publishers what they have to do to stop their content being used for Apple product development. 

Can you unmake an AI model?

What isn’t clear is the extent to which information already scraped by Applebot for use in Apple Intelligence (or any generative AI service) can then be winnowed out of the models Apple has already made. Once the model is created using your data, to what extent can your data be subsequently removed from it? The learning — and potential for copyright abuse — has already been baked in.

But where is the compensation for those who’ve made their knowledge available online? 

In most cases, the AI firms argue that what they are doing can be seen as fair use rather than being any violation of copyright laws. But, given that what constitutes fair use differs in different nations, it seems highly probable that the evolving AI industry is heading directly toward regulatory and legal challenges around their use of content.

That certainly seems to be part of the concern coming from regulators in some jurisdictions, and we know the legal framework around these matters is subject to change. This might also be part of what has prompted Apple to say it will not introduce the service in the EU just yet.

Move fast and take things

Right now, AI companies are racing faster than government regulation. Some in the space are attempting to side-step such debates by placing constraints around how data is trained. Adobe, for example, claims to train its imaging models only using legitimately licensed data. 

In this case, that means Adobe Stock images licensed content and older content that is outside of copyright.

Adobe isn’t just being altruistic in this — it knows customers using its generative AI (genAI) tools will be creating commercial content and recognizes the need to ensure its customers don’t end up being sued for illegitimate use of images and other creative works. 

What about privacy?

But when it comes to Apple Intelligence, it looks like the data you’ve published online has now become part of the company product, with one big exception: private data.

“We never use our users’ private personal data or user interactions when training our foundation models, and we apply filters to remove personally identifiable information like social security and credit card numbers that are publicly available on the Internet,” it said. 

Apple deserves credit for its consistent attempts to maintain data privacy and security, but perhaps it should develop a stronger and more public framework toward the protection of the creative endeavors of its customer base.

Please follow me on Mastodon, or join me in the AppleHolic’s bar & grill and Apple Discussions groups on MeWe.

The AI landscape is undergoing a transformative shift as chipmakers, traditionally focused on hardware innovation, are increasingly recognizing the pivotal role of software.

This strategic shift is redefining the AI race, where software expertise is becoming as crucial as hardware prowess.

AMD’s recent acquisitions: a case study

AMD’s recent acquisition of Silo AI, Europe’s largest private AI lab, exemplifies this trend. Silo AI brings to the table a wealth of experience in developing and deploying AI models, particularly large language models (LLMs), a key area of focus for AMD.

This acquisition not only enhances AMD’s AI software capabilities but also strengthens its presence in the European market, where Silo AI has a strong reputation for developing culturally relevant AI solutions.

“Silo AI plugs important capability gap [for AMD] from software tools (Silo OS) to services (MLOps) to helping tailor sovereign and open source LLMs and at the same time expanding its footprint in the important European market,” said Neil Shah, partner & co-founder at Counterpoint Research.

AMD’s move follows its previous acquisitions of Mipsology and Nod.ai, further solidifying its commitment to building a robust AI software ecosystem. Mipsology’s expertise in AI model optimization and compiler technology, coupled with Nod.ai’s contributions to open-source AI software development, provides AMD with a comprehensive suite of tools and expertise to accelerate its AI strategy.

“These strategic moves strengthen AMD’s ability to offer open-source solutions tailored for enterprises seeking flexibility and interoperability across platforms,” said Prabhu Ram, VP of industry research group at Cybermedia Research. “By integrating Silo AI’s capabilities, AMD aims to provide a comprehensive suite for developing, deploying, and managing AI systems, appealing broadly to diverse customer needs. This aligns with AMD’s evolving market position as a provider of accessible and open AI solutions, capitalizing on industry trends towards openness and interoperability.”

Beyond AMD: A broader industry trend

This strategic shift towards software is not limited to AMD. Other chip giants like Nvidia and Intel are also actively investing in software companies and developing their own software stacks.

“If you look at the success of Nvidia, it is driven not by silicon but by software (CUDA) and services (NGC with MLOps, TAO, etc.) it offers on top of its compute platform,” Shah said. “AMD realizes this and has been investing in building software (ROCm, Ryzen Aim, etc.) and services (Vitis) capabilities to offer an end-to-end solution for its customers to accelerate AI solution development and deployment.”

Nvidia’s recent acquisition of Run:ai and Shoreline.io, both specializing in AI workload management and infrastructure optimization, also underscores the importance of software in maximizing the performance and efficiency of AI systems.

But this doesn’t mean chipmakers follow similar trajectories toward their goals. Manish Rawat, semiconductor analyst at Techinsights pointed out that for a large part, Nvidia’s AI ecosystem has been established through proprietary technologies and a robust developer community, giving it a strong foothold in AI-driven industries.

“AMD’s approach with Silo AI signifies a focused effort to expand its capabilities in AI software, positioning itself competitively against Nvidia in the evolving AI landscape,” Rawat added.

Another relevant example in this regard is Intel’s acquisition of Granulate Cloud Solutions, a provider of real-time continuous optimization software. Granulate assists cloud and data center clients in optimizing compute workload performance while lowering infrastructure and cloud expenses.

Software to drive differentiation

The convergence of chip and software expertise is not just about catching up with competitors. It’s about driving innovation and differentiation in the AI space.

Software plays a crucial role in optimizing AI models for specific hardware architectures, improving performance, and reducing costs. Eventually, software could decide who rules the AI chip market.

“The bigger picture here is that AMD is obviously competing with NVIDIA for supremacy in the AI world,” said Hyoun Park, CEO and chief analyst at Amalgam Insights. “Ultimately, this is not just a question of who makes the better hardware, but who can actually back the deployment of enterprise-grade solutions that are high-performance, well-governed, and easy to support over time. And although Lisa Su and Jensen Huang are both among the absolute brightest executives in tech, only one of them can ultimately win this war as the market leader for AI hardware.” 

The rise of full-stack AI solutions

The integration of software expertise into chip companies’ offerings is leading to the emergence of full-stack AI solutions. These solutions encompass everything from hardware accelerators and software frameworks to development tools and services.

By offering a comprehensive suite of AI capabilities, chipmakers can cater to a wider range of customers and use cases, from cloud-based AI services to edge AI applications.

For instance, Silo AI, first and foremost, brings an experienced talent pool, especially working on optimizing AI models, tailored LLMs, and more, according to Shah. Silo AI’s SIloOS particularly is a very powerful addition to AMD’s offerings allowing its customer to leverage advanced tools and modular software components to customize AI solutions to their needs. This was a big gap for AMD.

“Thirdly, Silo AI also brings in MLOps capabilities which are a critical capability for a platform player to help its enterprise customers deploy, refine and operate AI models in a scalable way,” Shah added. “This will help AMD develop a service layer on top of the software and silicon infrastructure.”

Implications for enterprise tech

The shift of chipmakers from purely hardware to also providing software toolkits and services has significant ramifications for enterprise tech companies.

Shah stressed that these developments are crucial for enabling enterprise and AI developers to fine-tune their AI models for enhanced performance on specific chips, applicable to both training and inference phases.

This advancement not only speeds up product time-to-market but also aids partners, whether they are hyperscalers or manage on-premises infrastructures, in boosting operational efficiencies and reducing total cost of ownership (TCO) by improving energy usage and optimizing code.

“Also, it’s a great way for chipmakers to lock these developers within their platform and ecosystem as well as monetize the software toolkits and services on top of it. This also drives recurring revenue, which chipmakers can reinvest and boost the bottom line, and investors love that model,” Shah said.

The future of AI: a software-driven landscape

As the AI race continues to evolve, the focus on software is set to intensify. Chipmakers will continue to invest in software companies, develop their own software stacks, and collaborate with the broader AI community to create a vibrant and innovative AI ecosystem.

The future of AI is not just about faster chips — it’s about smarter software that can unlock the full potential of AI and transform the way we live and work.

Microsoft released 132 updates in its July Patch Tuesday update while addressing four zero-days (CVE-2024-35264, CVE-2024-37985, CVE-2024-38080 and CVE-2024-38112) affecting Windows desktop, Microsoft .NET and Visual Studio. This is a very significant patch cycle for Microsoft SQL Server, but there are no updates for Microsoft browsers and a low profile set of patches for Microsoft Office. No major revisions require attention, with testing focused squarely on SQL dependent applications. 

The team at Readiness has provided a useful infographic detailing the risks with each of the updates this cycle. 

Known issues 

Each month, Microsoft publishes a list of known issues included in its latest release, including two reported minor issues:

• After you install KB5034203 (dated 01/23/2024) or later updates, some Windows devices that use the DHCP Option 235 to discover Microsoft Connected Cache (MCC) nodes in their network might be unable to use those nodes. Microsoft offered two options to mitigate the issue through setting the Cache Hostname or using group policies. Microsoft is still working on a resolution.
• Context menus and dialog buttons in some Windows apps, or parts of the Windows OS user interface (UI), might display in English when English is not set as the display language. This might also affect font size.

We fully expect to see more issues relating to how the Windows UI presented over the coming months as Microsoft works through some of the core level issues with new ARM builds. This means that even non-ARM builds will be affected (see CVE-2024-37985). Look out for input method editor, language pack, and dialog box language issues for non-English builds.

Major revisions 

This Patch Tuesday saw Microsoft publishing the following major revisions to past  security and feature updates, including:

• CVE-2024-30098 : Windows Cryptographic Services Security Feature Bypass. Microsoft has added a FAQ to explain how this vulnerability is being addressed and further actions customers must take to be protected from it. This is an informational change only; no further action is required.

Mitigations and workarounds

Microsoft published the following vulnerability-related mitigations for this month’s release cycle: 

• CVE-2024-38077, CVE-2024-38076 and CVE-2024-38074: Windows Remote Desktop Licensing Service Remote Code Execution Vulnerability. Microsoft has (rather helpfully) offered guidance that disabling the Remote Desktop licensing feature mitigates this vulnerability. 
• CVE-2024-38030: Windows Themes Spoofing Vulnerability. If you have Windows NTLM disabled, this vulnerability will not affect your system.

Each month, the Readiness team analyses the latest Patch Tuesday updates and provides detailed, actionable testing guidance based on assessing a large application portfolio and a detailed analysis of the patches and their potential impact on the Windows platforms and app installations.

For this cycle, we have grouped the critical updates and required testing efforts into different functional areas:

Microsoft Office

• Test out your Teams logins (which shouldn’t take too long).
• Because SharePoint was updated, third-party extensions or dependencies will require testing.
• Due to the change in Outlook, Internet Calendars (ICS files) will require testing.
• With the Visio update, large CAD drawings will require a basic import and load test.

Microsoft .NET and developer tools

Microsoft has updated the Microsoft .NET, MSI Installer and Visual Studio with the following testing guidance:

• PowerShell updates will require a diagnostics test. Try the command, “import-module Microsoft.powershell.diagnostics – verbose” and validate that you are getting the correct results from your home directory.
• Due to the change in the Windows core installation technology (MSI), please validate that User Account Control (UAC) still functions as expected.

Microsoft SQL Server

This month is a big update for both Microsoft SQL Server and the local, or workstation supporting elements of OLE. The primary focus for this kind of complex effort should be your line-of-business or core applications. These are the applications that have multiple data connections and rely on complex, multiple object/session requirements. Due to the changes this month, we can’t recommend specific Windows feature testing regimes, as we are most concerned that the business logic (and resulting data) of the application in question might be affected. Only you will know what looks good; we advise a comparative testing regime across unpatched and newly patched systems looking for data disparities.

Windows

Microsoft made another update to the Win32 and GDI subsystems with a recommendation to test out a significant portion of your application portfolio. We also recommend that you test the following functional areas in the Windows platform:

• File compression has been updated, so file and archive extraction scenarios will need to be exercised.
• Due to the Microsoft codec updates, perform a system reboot and test that your audio and camera still work together.
• Security updates will require the testing of the creation of new Windows certificates.
• Networking changes will require a test of DNS and DHCP, specifically the DHCP R_DhcpAddSubnetElement API. As part of these changes, testing VPN authentication will be required. Try to include your Network Policy Server (NPS) as part of the connection creation and deletion effort.
• This month’s update to Remote Desktop Services (RDS) will require the creation and revocation of license requests.
• A significant update to the Network Driver Interface Specification (NDIS) will require testing of network traffic involving repeated bursts of large files. Try using Teams while this networking burst testing is in progress.
• Backup and printing have been updated, so test your volumes and ensure that when you print out a test page, your OS does not crash (yes, really). Try printing out TIFF files. (Hey, you might like it.)

As part of the ongoing effort to support the new ARM architecture, Microsoft released the first patch for this new platform, CVE-2024-37985. This is an Intel assigned processor-level vulnerability that has been mitigated by a Microsoft OS level patch. The Readiness team has provided guidance on potential ARM-related compatibility and testing issues. 

Specifically, the Readiness team was concerned with Input Method Editors (IMEs). We suggest a full test cycle of Windows input related features such as keyboard, mouse, touch, pen, gesture and dictation. Some internet shortcuts might be affected as well as wallpapers.

Windows lifecycle update 

This section contains important changes to servicing (and most security updates) to Windows desktop and server platforms.

• Home and Pro editions of Windows 11, version 22H2 will reach end of service on Oct. 8, 2024. Until then, these editions will only receive security updates. They will no longer receive non-security, preview updates.

Each month, we break down the update cycle into product families (as defined by Microsoft) with the following basic groupings: 

• Browsers (Microsoft IE and Edge);
• Microsoft Windows (both desktop and server); 
• Microsoft Office;
• Microsoft Exchange Server ;
• Microsoft Development platforms (ASP.NET Core, .NET Core and Chakra Core)
• Adobe (if you get this far).

Browsers

Microsoft did not release any updates for its non-Chromium browsers. Following the stable channel release of Chrome (applicable until July 25, 2024) we have not seen any changes, deprecations or testing profile updates to this browser. No further action required.
 

Windows

Microsoft released four critical and 83 updates rated as important with two zero-day patches (CVE-2024-38080 and CVE-2024-38112) affecting the Microsoft Hyper-V and MSHTML feature groups, respectively. In addition to these critical updates, Microsoft patches for July affect the following Windows feature groups:

• Windows NTLM, Kernel, GDI and Graphics;
• Windows Backup;
• Windows Codecs;
• Microsoft Hyper-V;
• Windows (Line) Print and Fax ;
• Windows Remote Desktop and Gateway;
• Windows Secure Boot and Enrolment Manager.

Add these Windows updates to your Patch Now release cycle.

Microsoft Office 

Microsoft returns to form with a critical update for Office this month (CVE-2024-38023) for the SharePoint platform. We have another update for Outlook related to spoofing (CVE-2024-38020), but this vulnerability is not wormable and requires user interaction. There are four more, lower rated updates; please add all of these updates to your standard release schedule.

Microsoft SQL (nee Exchange) Server 

There were no updates for Microsoft Exchange Server this month. However, we have seen the largest release of Microsoft SQL updates in the past few years. These SQL-related updates cover 37 separate reported vulnerabilities (CVEs) and the following main product features

• SQL Server Native Client OLE DB Provider;
• Microsoft OLE DB Driver for SQL.

We covered the testing requirements for this SQL update in our testing guidance section above. This month’s SQL updates will require some preparation and dedicated testing before adding to your standard release schedule.

Microsoft development platforms 

Microsoft released four, low-profile updates to the Microsoft .NET and Visual Studio platforms. We do not expect serious testing requirements for these vulnerabilities. However, CVE-2024-35264 has been reported as publicly disclosed by Microsoft. This makes this an unusually urgent patch for Microsoft Visual Studio attracting a “Patch Now” rating this month.

Adobe Reader (and other third-party updates) 

Very much as our Microsoft Exchange section has been “hijacked” by SQL Server updates this month, we’re using the Adobe section for third-party updates. (There are no updates to Adobe Reader.) 

• CVE-2024-3596: NPS RADIUS Server. A vulnerability exists in the RADIUS protocol that potentially affects many products and implementations of the RFC 2865 in the UDP version of the RADIUS protocol. 
• CVE-2024-38517 and CVE-2024-39684: GitHub Active Directory Management Rights. The  vulnerability assigned to this CVE is in the RapidJSON library which is consumed by the Microsoft Active Directory Rights Management Services Client, hence the inclusion of this CVE with this update.
• CVE-2024-37985: This memory related update from Intel relates to its prefetcher technology. Affected: Core Windows OS memory related components — particularly the new ARM builds, which I find both confusing and ironic.

The European Commission has released the preliminary findings from an investigation launched last year into X (formerly Twitter), and said it believes the company is in breach of the Digital Services Act (DSA), which applies to marketplaces, social networks, content-sharing platforms, app stores, and online travel and accommodation platforms.

Non-compliance in three areas

In a statement, the Commission said X was found non-compliant in three areas: 

• The “verified account” mechanism is designed and implemented in a way that deceives users and does not correspond to industry practice. “Since anyone can subscribe to obtain such a ‘verified’ status, it negatively affects users’ ability to make free and informed decisions about the authenticity of the accounts and the content they interact with,” the Commission said, adding there is “evidence of motivated malicious actors abusing the ‘verified account’ to deceive users.”
• X does not comply with requirements around transparency in advertising. “In particular, the design does not allow for the required supervision and research into emerging risks brought about by the distribution of advertising online,” the Commission  argued.
• X does not provide access to its public data to researchers, as specified by conditions in the DSA. Its terms of service prohibit researchers from independently accessing public data, and its process for granting researchers access via its application programming interfaces (APIs) “appears to dissuade researchers from carrying out their research projects or leave them with no other choice than to pay disproportionally high fees.”

X now has the right to examine the commission’s documentation and prepare a defense. 

If the preliminary findings are confirmed, the company faces a non-compliance decision that could result in fines of up to 6% of its global annual revenue, an order to address the issues detailed in the decision, and the potential for a period of enhanced supervision. The commission  can also impose periodic penalty payments.

The move could be seen as a warning shot to other companies.

“While the ruling may not have a direct impact on enterprise CIOs, it emphasizes learning from broader implications and the mistakes of others,” said Phil Brunkard, executive counselor at Info-Tech Research Group, UK. “It sets a precedent for public trust in online marketplaces or social media, highlighting the importance of integrity and transparency in data privacy. Regulation is not just about ticking the compliance box — it’s crucial for customer trust. CIOs must ensure strong governance to protect their brands and maintain customer trust, as trust is the foundation for successful organizations.”

Investigations continue

Investigations continue into X’s risk management around the dissemination of illegal content and the effectiveness of how it combats information manipulation.

To assist in its investigations, the Commission released a whistleblower tool that allows people to contact it anonymously with information contributing to compliance monitoring of X and other entities designated Very Large Online Platforms (VLOP) under the DSA.

X is not the only organization under scrutiny. The Commission has also initiated formal proceedings against TikTok, Meta (in separate proceedings launched in April and May 2024, respectively), and AliExpress.