Amodei's Six-Month Cyber Countdown.

TL;DR

• Cybersecurity Imperative: Anthropic CEO Dario Amodei issued a 6-to-12-month timeline for organizations to patch "tens of thousands" of software vulnerabilities identified by Mythos, underscoring the urgency before Chinese AI capabilities close the gap. This critical warning was delivered alongside the launch of ten new financial services agents and the announcement of Anthropic's $30 billion revenue run-rate, now surpassing OpenAI. (CNBC, Fortune, Bloomberg)
• Strategic Compute Access: Anthropic secured full compute capacity at SpaceX's Colossus 1 data center, encompassing over 220,000 Nvidia GPUs and 300+ megawatts, to address escalating Claude Code demand. This unexpected collaboration follows Elon Musk's previous public criticisms, now softened as he expressed an "impressed" stance after engaging with Anthropic's team. Orbital AI compute exploration is also underway between the entities. (Reuters, CNBC, Anthropic)
• OpenAI's Dual Thrust: OpenAI rolled out GPT-5.5 Instant as ChatGPT's default, touting a 52.5% reduction in hallucinations for critical prompts. Concurrently, it launched a self-serve advertising platform for all U.S. businesses, introducing cost-per-click bidding and signaling a significant shift in its monetization strategy. (OpenAI Blog, Axios)
• Universal Government Oversight: All major U.S. frontier AI labs, including Google DeepMind, Microsoft, and xAI, have now established formal agreements for pre-release government evaluation via CAISI, aligning with existing OpenAI and Anthropic commitments. This consolidation of oversight precedes potential mandatory executive action from the White House. (Politico, Reuters)
• Musk v. Altman & DeepSeek's Valuation: The Musk v. Altman trial continued with Shivon Zilis's testimony, following Greg Brockman's disclosures regarding Musk's demand for 51% of OpenAI for Mars city funding and the exponential growth of compute budgets. Separately, DeepSeek's inaugural fundraise is reportedly nearing a $45 billion valuation, led by China's state "Big Fund," highlighting critical shifts in global AI investment and control. (NBC News, Financial Times, TechCrunch)

Lead Story: Amodei's Alarm and a New AI-Finance Alliance

The week's most consequential AI development transpired not within a courtroom or via a product unveiling, but on a Lower Manhattan stage, where Anthropic CEO Dario Amodei shared a panel with JPMorgan Chase CEO Jamie Dimon. This initial joint appearance underscored the financial sector's direct engagement with frontier AI capabilities.

The context was Anthropic's "Briefing: Financial Services" event, showcasing ten new purpose-built AI agents for banking and back-office functions, alongside comprehensive Microsoft 365 integration and strategic data partnerships with Moody's, Third Bridge, and Dun & Bradstreet. Amodei concurrently revealed Anthropic's annualized Q1 2026 revenue surged approximately 80x, significantly exceeding internal projections. This trajectory propelled Anthropic's April run-rate revenue beyond $30 billion, eclipsing OpenAI's $25 billion. Over 1,000 enterprise clients now commit over $1 million annually to Claude, a figure that doubled in less than two months, evidencing rapid market penetration.

Amodei's cybersecurity warning was unequivocal. Mythos, Anthropic's restricted frontier model, identified nearly 300 Firefox vulnerabilities, a drastic increase from the roughly 20 discovered by its predecessor three months prior. This extends to "tens of thousands" of total vulnerabilities across sectors. The critical window for remediation: six to twelve months, reflecting the estimated lead Mythos holds over comparable Chinese models. Amodei articulated the risk as "an enormous increase in the amount of vulnerabilities, in the amount of breaches, in the financial damage that's done from ransomware on schools, hospitals, not to mention banks," per CNBC, framing it as a systemic economic threat.

Dimon confirmed cybersecurity as JPMorgan's "biggest risk for years," positing Mythos as a transformative development. While OpenAI CEO Sam Altman previously dismissed Anthropic's public warnings as "fear-based marketing," Amodei, without direct rebuttal, advocated for "legislation, or rules" governing powerful AI model releases. This signals a notable escalation in Anthropic's policy position.

This event also provided tangible product validation for the recently announced Anthropic-Blackstone-Goldman Sachs-Hellman & Friedman joint venture: the new financial services agent suite aligns precisely with the $1.5 billion JV's deployment strategy for private equity portfolio companies. An internal memo from OpenAI's chief revenue officer, however, disputed Anthropic's $30 billion revenue claim, citing an approximate $8 billion overstatement due to revenue recognition discrepancies. Anthropic maintains its pricing autonomy and principal role in these transactions. This accounting contention underscores the escalating competitive intensity as both entities vie for the same high-value enterprise accounts. (CNBC, Fortune, Bloomberg, Anthropic, The Register)

In Other News

Every Major Lab Now Under Government Review — and an EO May Make It Mandatory. The Commerce Department's CAISI announced pre-deployment evaluation agreements with Google DeepMind, Microsoft, and xAI, finalizing a comprehensive framework encompassing all major U.S. frontier labs. OpenAI and Anthropic had previously recalibrated their Biden-era agreements to align with current administration directives. CAISI has now conducted over 40 model evaluations, including unreleased systems, with developers reportedly providing models sans safety guardrails for enhanced national security scrutiny. Politico confirms the White House is drafting an executive order to mandate such government vetting, potentially establishing an AI working group. While current agreements are voluntary, the proposed EO would be compulsory. The Trump administration, despite revoking Biden's prior executive order, has now instituted a more robust pre-deployment oversight apparatus without legislative action. (Politico, Reuters, Axios, CNN)

OpenAI Launches GPT-5.5 Instant as Default — and Opens the Ad Store. OpenAI executed a dual-pronged product offensive. GPT-5.5 Instant is now the default ChatGPT model for all users, including the free tier, with claims of a 52.5% reduction in hallucinations for high-stakes medical, legal, and financial prompts, alongside improved image reasoning and STEM benchmarks. The model also leverages past chats, files, and Gmail for personalized responses, managed via explicit "memory sources" controls. More strategically, OpenAI has opened its self-serve ChatGPT Ads Manager to all U.S. businesses, introducing cost-per-click bidding to supplement existing CPM models. The company projects $2.5 billion in ad revenue this year, targeting $100 billion by 2030. GPT-5.5 Instant represents an incremental quality enhancement; the ad platform signifies a fundamental business model transformation, intensifying the inherent tension between user personalization and data privacy. (OpenAI Blog, Axios, Digiday)

Trial Days 6-7: Brockman Wraps, Zilis Takes the Stand. Greg Brockman concluded his two-day testimony in Musk v. Altman with notable disclosures, including Musk's demand for a 51% controlling stake in OpenAI to fund an $80 billion Mars city. Brockman recounted Musk becoming physically threatening upon rejection and revealed OpenAI's compute budget escalated from $30 million in 2017 to $50 billion in 2026. He also stated Musk directed OpenAI employees to covertly work on Tesla's self-driving technology, characterizing his self-published journal entries as "raw, unfiltered thoughts" never intended for public scrutiny. Shivon Zilis's testimony commenced Wednesday afternoon. A Neuralink executive, mother to four of Musk's children, and former OpenAI board member, Zilis's role from 2020-2023 raises the central question, as framed by Vanity Fair: whether she functioned as an "unwitting agent" providing Musk a back channel into OpenAI's governance during a pivotal period. Her testimony is scheduled to continue, as Musk pursues $150 billion in damages and the removal of Altman and Brockman. (CNBC, NBC News, The Guardian, Vanity Fair)

Musk's Supercomputer, Anthropic's Gain. In an unexpected strategic alignment, Anthropic announced an agreement with SpaceX to utilize the full compute capacity of Colossus 1, the massive Memphis data center originally built by xAI and acquired by SpaceX. This facility comprises over 220,000 Nvidia GPUs, including H100, H200, and GB200 accelerators, with a total capacity exceeding 300 megawatts. Elon Musk, a recent vocal critic of Anthropic, adopted a notably different tone on X, citing engagement with Anthropic's senior team and expressing his "impressed" stance. He reserved the right to reclaim compute if Anthropic's AI "engages in actions that harm humanity," an unprecedented clause he justified by stating SpaceXAI's training had migrated to Colossus 2. Anthropic also expressed interest in partnering with SpaceX on multi-gigawatt orbital AI compute capacity, a critical objective ahead of SpaceX's IPO. Concurrently, Anthropic unveiled "dreaming," a self-improvement feature enabling Claude agents to review their work between sessions. This deal represents pure operational pragmatism: Anthropic gains essential GPU resources for escalating Claude Code demand, while SpaceX secures a prominent customer for its IPO narrative. (Reuters, CNBC, Bloomberg, Anthropic)

X / Social Pulse

The SpaceX-Anthropic compute agreement profoundly shaped Wednesday's discourse, eclipsing Amodei's cyber warning. The inherent irony was stark: Musk is concurrently litigating against OpenAI (founded by Anthropic's principals), publicly recalibrating his disposition toward Anthropic, and provisioning his supercomputing assets to them, all while his civil trial progresses through critical testimony. Security researchers observed that Musk's "right to reclaim compute" clause is without precedent in data center leasing and likely legally tenuous, serving more as a public signaling mechanism than an enforceable contractual term.

Amodei's 6-to-12-month remediation window fueled considerable debate. Leading researchers challenged the premise that Chinese AI would require such a timeframe to achieve Mythos-level capabilities, citing prior replication of specific cyberattack functionalities via extant open-source tools. His advocacy for "legislation" regarding frontier model releases represents a distinct departure from Anthropic's prior circumspect policy posture.

The confluence of ChatGPT advertisements and persistent memory features initiated an immediate privacy discourse. The $3-$5 CPC recommendation draws parallels to Google Ads in its early 2004 phase: nascent, possessing immense market potential, yet with an underdeveloped understanding of conversational intent signals. OpenAI now functions as a productivity tool, an advertising platform, and a personal data repository. The inherent tension across these operational roles will undoubtedly intensify.

One to Watch

DeepSeek's First Fundraise — State-Backed and Geopolitically Loaded. Reports from the FT, corroborated by Reuters, Bloomberg, and TechCrunch, indicate China's "Big Fund" — the state's principal semiconductor investment vehicle — is poised to lead DeepSeek's initial external funding round at a $45 billion valuation, a substantial escalation from April's $10 billion figure. Tencent and Alibaba are also engaged in discussions. DeepSeek's prior self-funding model renders this state-backed, conglomerate-reinforced investment a critical signal: DeepSeek is being designated as national infrastructure, not simply a venture-backed startup. This aligns directly with Amodei's 6-to-12-month cybersecurity warning, positioning a nationally prioritized DeepSeek as a key instrument in China's AI catch-up timeline. A recent AP and LA Times investigation further revealed China's emergence as "a testing ground for mass use of AI tools," with adoption rates exceeding the U.S. in specific sectors. (Financial Times, Reuters, TechCrunch, AP News)

Quick Hits

• Robotics Foundation Model Advancement: Ai2 introduced MolmoAct 2, a fully open-source robotics foundation model demonstrating superior performance to Physical Intelligence's pi0.5 in both simulation and zero-shot real-world applications. It also boasts a 37x speed improvement and includes an open-source action tokenizer.
• AWS-OpenAI Integration: AWS confirmed limited preview access for GPT-5.5 and GPT-5.4 on Amazon Bedrock, alongside Codex and OpenAI Managed Agents, signifying a deepening of the AWS-OpenAI platform partnership.
• Landmark State AI Legislation: Connecticut's SB 5, a comprehensive state AI bill addressing frontier models, chatbots, employment AI, and youth social media, has cleared the legislature. Governor Lamont is expected to sign it into law.
• Apple's Siri Settlement: Apple agreed to a $250 million settlement concerning claims of misleading iPhone consumers regarding Siri's AI capabilities. Approximately 36 million eligible device owners may receive $25 to $95 each. (NYT, The Guardian)
• Pervasive AI Vulnerabilities: A scan of over 2 million AI hosts identified approximately 1 million exposed services operating without authentication or adequate safety controls. This widespread unsecure self-hosting of LLM infrastructure generates a substantial, searchable attack surface, directly validating the risk profile outlined by Amodei. (The Hacker News)

Wednesday's events underscored a profound reordering of market dynamics. Musk navigated a courtroom battle for industry control in the morning, only to provision his supercomputing assets to a former adversary by afternoon. Government oversight expanded to all major AI labs, preceding a fully defined regulatory framework. OpenAI initiated a critical business model shift via its advertising platform. Concurrently, Anthropic, now a revenue leader, issued a stark warning: the window for responsible action is measured in months, not years. The industry's prior equilibrium has dissolved. Its successor is being forged concurrently across legal, infrastructural, and commercial fronts, driven by agents operating with acute strategic improvisation.

Sources

CNBC | Reuters | Fortune | Axios | New York Times | NBC News | Business Insider | Wired | Politico | The Verge | Financial Times | Bloomberg | OpenAI Blog | Anthropic Blog | Digiday | The Guardian | Vanity Fair | TechCrunch | AP News | CNN | The Register | The Hacker News | LA Times

Lock in. M. mazen@thorterminal.com