Last week gave us Nvidia's GTC, Amazon's robotics acquisition, and the White House's AI framework. This week the spotlight shifts from big product announcements to harder questions: how do you secure AI systems that nobody fully controls, and who gets to set the rules for what they can do?
Here is what to watch from Monday.
Apple's iOS 26.4 beta should drop this week
Apple has been promising a rebuilt Siri for two years. The first beta of iOS 26.4 is expected to arrive this month, and with it, the clearest signal yet of what the company's AI overhaul actually delivers, according to MacRumors.
The backstory is messy. The new version of Siri could arrive with iOS 26.4 this spring or be pushed to iOS 27 in September, Information Age has reported. Reports surfaced in February that some features were not behaving as intended during testing, raising questions about how much of the promised overhaul ships this spring versus how much gets deferred to the autumn.
Apple is working on an internal project codenamed "Campos" that will transform Siri into the company's first AI chatbot, running on a version of Google's Gemini model internally referred to as Apple Foundation Models version 11, Business Standard reported. The deal between Apple and Google, confirmed in January, is broader than just Siri: it covers Apple's entire intelligence layer going forward.
Some features will launch this spring, while others, including Siri's ability to remember past conversations and proactive calendar-aware suggestions, are expected to be announced at WWDC in June, 9to5Mac reported. This beta release will clarify which bucket most of the headline features land in. If the gap between what Apple teased at WWDC 2024 and what actually ships in 26.4 is large, expect the stock to respond accordingly.
RSAC takes on agentic AI, for the first time at scale
The cybersecurity world's most important gathering kicks off Monday at San Francisco's Moscone Center and runs through Thursday. RSAC 2026 marks its 35th anniversary, having grown from a single panel of 50 cryptographers in 1991 to more than 43,000 attendees, according to GovInfoSecurity. The theme this year is "Power of Community," but the substance is almost entirely about AI.
Programming will spotlight agentic artificial intelligence, incident response, and responsible disclosure, GovInfoSecurity reported. That framing matters. Last year, most AI security conversations were theoretical. This year, organisations are actually deploying agents in production, and the attack surface has expanded to match.
Microsoft's Monday keynote covers how ambient, autonomous platforms are evolving to address AI-powered threats. Expect similar territory from CrowdStrike, Google, and a packed floor of vendors competing to define what AI-native security actually looks like in practice.
One session worth noting: on Thursday, security veterans Kevin Mandia and Nicole Perlroth deliver a keynote examining the past year's most consequential cyberattacks and the threats ahead, covering AI-automated attacks, insecure LLM-generated code, and evolving nation-state tactics, according to PR Newswire. Their framing, that "the AI honeymoon in cybersecurity may be ending," signals a mood shift from last year's optimism.
One notable absence: US federal agencies have declined to participate this year, GovInfoSecurity reported, leaving a gap that international security leaders from the UK, UAE, and European Commission will fill.
The AI copyright wars move to a new front
Encyclopedia Britannica and Merriam-Webster filed a lawsuit against OpenAI, alleging the company scraped nearly 100,000 copyrighted articles to train its models without permission, Neural Buddies reported. The case has a wrinkle that earlier suits did not: it also targets OpenAI's retrieval-augmented generation workflow for allegedly reproducing Britannica content when generating responses, one of the first major lawsuits to target this specific AI technique.
That matters beyond the two plaintiffs. RAG is the backbone of how most enterprise AI systems work, pulling from databases and documents at query time rather than relying solely on training data. If a court finds that RAG outputs constitute copyright infringement, the implications for the industry would be significant, and far messier to unpick than the original training-data cases.
AI's upcoming legal battles are expected to be far more complex than the earlier wave of training-data suits, MIT Technology Review has argued, centring on thorny unresolved questions: whether AI companies can be held liable for what their chatbots encourage people to do, and whether insurers will shun AI companies as clients if they start losing cases. The Britannica suit adds another dimension to a legal picture that is growing more complicated by the month.
The Transform Conference asks who pays the workforce cost
While security leaders convene in San Francisco, a different set of conversations will be happening 400 miles to the south-east. The Transform Conference 2026, themed "The Human and AI Equation: Forging the Next Era of Work," runs Monday through Wednesday in Las Vegas, attracting CHROs, heads of people operations, and senior leaders navigating the human implications of AI adoption, according to ALM Corp.
Transform tends to be where the more uncomfortable conversations happen. The RSAC crowd is focused on securing AI systems. The Transform crowd is focused on what those systems mean for the people who work alongside them, or used to. In a week where Amazon's robotics acquisition is still fresh and the White House's AI framework has barely dried, the timing is pointed.
AI advertising hits its first serious reality check
One story that developed quietly last week deserves more attention heading into this one. One brand running ads inside ChatGPT pulled a click-through rate nearly seven times below Google search benchmarks in the same vertical, and another advertiser only managed to spend 3% of a $250,000 budget after several weeks because the platform could not push enough volume, The AI Marketers reported. OpenAI is also dealing with a reporting glitch inside its Ad Manager that is blocking brands from accessing their own data, leaving no real way to optimise performance.
This matters because AI advertising has been sold as the next great channel, with enormous sums moving in that direction. US advertisers are projected to push $57 billion through AI-powered platforms this year, a 63% increase, while the share still managed by humans is growing at just 5%, according to research cited by The AI Marketers. If the early ChatGPT ad numbers reflect the channel's ceiling rather than its floor, that is a significant problem for a company that needs advertising revenue to justify its valuation trajectory.
The financial results that matter from last week
The earnings calendar delivered some sharp signals before the weekend worth carrying into the week ahead.
Micron's results, reported Wednesday, were the standout. Revenue in the fiscal second quarter almost tripled year-on-year to $23.86 billion, against analyst expectations of $20.07 billion, with adjusted earnings per share of $12.20 against a consensus estimate of $9.31, CNBC reported. The guidance was even more striking: the company now expects around $33.5 billion in revenue for the current quarter, implying year-on-year growth of over 200%, with CEO Sanjay Mehrotra citing AI as the structural driver.
The memory market is benefiting from a specific dynamic. Each generation of Nvidia chip requires more memory, creating a supply crunch that Micron and its competitors Samsung and SK Hynix are racing to fill. For anyone tracking AI infrastructure spend, Micron is now one of the cleaner financial proxies for how fast the buildout is actually accelerating.
On the consumer side, Darden Restaurants offered a quieter but useful read on the broader economy. The Olive Garden and LongHorn Steakhouse parent posted Q3 revenue of $3.3 billion, up 5.9% year-on-year, driven by same-restaurant sales growth of 4.2%, according to its investor release. The company raised its full-year guidance, now projecting same-restaurant sales growth of around 4.5% and 70 new restaurant openings, RTT News reported, up from the 60 previously targeted. Sit-down dining holding up at that level, as quick-service prices remain elevated, is a worth watching as a consumer confidence indicator heading into Q2.
The week will not resolve any of these questions cleanly. But between RSAC's 43,000 attendees stress-testing AI security assumptions, Apple's first real signal on Siri, and courts beginning to define what AI systems are and are not allowed to do with other people's work, the next five days should tell us a fair amount about where the year is heading.