AI Step-Change Alert!

An inadvertent data exposure revealed that Anthropic has developed and is testing a new model internally referred to as both "Claude Mythos" and "Capybara." The leak, identified by an AI security researcher at LARX Security and a cybersecurity researcher at the University of Cambridge, consisted of draft blog posts and supporting documents stored in an unsecured, publicly searchable data cache. The leaked materials describe Capybara as a new tier of model—positioned above Anthropic's existing "Opus" tier, which had previously been its most capable offering. Anthropic's current model lineup runs from Haiku (smallest, fastest, cheapest) through Sonnet to Opus (largest, most capable). Capybara would represent an entirely new category above Opus, with the draft documentation citing dramatically higher benchmark scores in software coding, academic reasoning, and cybersecurity tasks. The cybersecurity implications are the most operationally significant element. The leaked draft states that the model is "far ahead of any other AI model in cyber capabilities" and that it "presages an upcoming wave of models that can explore and exploit vulnerabilities in ways that far outpace the efforts of defenders." Anthropic's stated release strategy is accordingly defensive-first: early access is being granted to organizations focused on hardening their code bases against AI-driven exploits, rather than a broad commercial rollout. This framing echoes OpenAI's February release of GPT-5.3 Codex, which the company classified as the first model meeting a "high capability" threshold for cybersecurity tasks under its internal preparedness framework. The disclosure also revealed plans for an invite-only CEO retreat at an English country manor, described as targeting Europe's most influential business leaders, with Anthropic CEO Dario Amodei attending. The event is positioned as part of a broader enterprise sales push. Separately, the discussion covers reporting that Anthropic executives have discussed a public offering as soon as Q4 of this year, with bankers expecting the company to raise more than $60 billion in an IPO.

Apple is planning to replace its exclusive ChatGPT integration within Siri with a broader "extensions" system in iOS 27, according to reporting attributed to Bloomberg's Mark Gurman. Under the new architecture, any AI service distributed through the App Store—including Google Gemini and Anthropic's Claude—could be invoked directly through Siri, provided the third-party developer enables the capability. The strategic logic is twofold. First, it eliminates the need for individually negotiated integration deals like the existing ChatGPT arrangement, allowing Apple to scale AI partnerships without repeated business negotiations. Second, it creates a new revenue stream: Apple would collect App Store commissions on third-party AI subscriptions initiated through Siri queries. The discussion notes an important distinction between this extensions layer and a separate, deeper arrangement in which Apple is working with Google to rebuild Siri's underlying technology using Gemini models. The extensions system operates at the user-facing request level; the Gemini infrastructure work operates at the platform level. These are parallel, not competing, initiatives. The announcement is expected at Apple's Worldwide Developers Conference on June 8th, though features could change or be delayed before then. In unrelated Apple news, the company has officially discontinued the Mac Pro line with no plans for future hardware in that category, effectively consolidating its high-end desktop offering around the Mac Studio.

OpenAI has surpassed $100 million in annualized advertising revenue from ChatGPT approximately six weeks after launching its ad pilot, according to a company spokesperson. The figure is notable for what it excludes: the revenue was generated from less than 20% of eligible US-based free and "Go" tier users. As of the reporting date, approximately 85% of free and Go users are eligible to see ads, suggesting significant headroom for growth without expanding the user base. The company has engaged more than 600 advertisers and plans to launch self-serve advertising access in April—a move that would lower the barrier to entry for smaller advertisers and accelerate inventory fill. OpenAI recently hired a former Meta advertising executive to lead ad sales and is evaluating geographic expansion into Canada, Australia, and New Zealand. The company has told investors it expects to generate more than $17 billion from consumer-facing ChatGPT products in 2026, with advertising representing a meaningful component of revenue from its non-paying user base.

SpaceX is reportedly planning to allocate as much as 30% of its IPO to individual retail investors—three to six times the typical 5–10% retail allocation in public listings. The plan, relayed to Wall Street by SpaceX CFO Brett Johnson, pairs the oversized retail tranche with a selective banker engagement model in which firms are assigned narrow roles based on personal relationships rather than open competition. Bank of America has reportedly been designated for domestic retail distribution. The structure reflects a bet that retail investors with long-term conviction in the company are less likely to engage in immediate post-listing selling. --- Key takeaways: - Anthropic's leaked Capybara/Mythos model introduces a new capability tier above its existing Opus line, with the company's own documentation flagging it as posing unprecedented cybersecurity risks—its release strategy prioritizes defensive security organizations over general commercial availability. - Enterprise security teams should treat the imminent arrival of AI models capable of large-scale automated vulnerability exploitation as a near-term operational threat, not a theoretical one, given Anthropic's own framing and OpenAI's parallel classification of Codex under its preparedness framework. - Apple's shift to an open AI extensions model in iOS 27 restructures the competitive dynamics of the AI assistant market, potentially benefiting Anthropic and Google while reducing ChatGPT's privileged position and generating new App Store revenue for Apple. - OpenAI's advertising business is scaling faster than most platform ad businesses at launch, with $100 million in annualized revenue from a fraction of eligible users suggesting the full rollout could materially alter the company's consumer revenue profile ahead of its anticipated IPO. - The SpaceX IPO's unusually large retail allocation signals a deliberate strategy to build a stable long-term shareholder base, with demand expectations drawing comparisons to Google's 2004 public offering.