US government suspends Anthropic's Fable 5 and Mythos 5 models

US Commerce Department restricts Anthropic's Fable and Mythos models to US citizens after Amazon researchers discover jailbreak

Following a vulnerability discovered by Amazon researchers who extracted security details and reported findings to the U.S. government rather than through responsible disclosure, Commerce Secretary Howard Lutnick imposed export restrictions on Anthropic's newly released Fable 5 and Mythos 5 models. The order prevents Anthropic from offering these models to non-US citizens, including its own international employees, and forced the company to withdraw both models globally. This represents a concrete demonstration of the gap between Anthropic's safety rhetoric and actual model robustness, with direct business consequences including potential IPO damage and competitive disadvantage versus other frontier labs continuing to ship. The incident also signals escalating tensions between defense officials and Anthropic, and comes as the company publicly signals support for pausing frontier AI development alongside OpenAI and DeepMind.

Tokenized SpaceX equity launches with mixed results: on-chain platforms execute cleanly while centralized exchanges fail to source shares

SpaceX opened for tokenized trading Friday with Ondo's SPCXon on Solana and Hyperliquid's pre-IPO perpetual executing cleanly, but Bybit, Binance, and Bitget canceled their tokenized allocations after xStocks failed to source underlying shares. The divergence exposes both the viability of on-chain equity tokenization and the operational challenges of bridging traditional share custody with crypto rails. The week also saw Digital Asset close a $355 million Series C led by a16z crypto for its Canton Network, valued above $2 billion with participants including HSBC, Apollo, CME, BNP Paribas, ADIA, S&P Global, and Tradeweb. Canton generated $60.74 million in fees over the past 30 days—ahead of Tron and Ethereum over the same window—signaling accelerating institutional adoption of tokenized settlement infrastructure.

Microsoft reaches 46.8% operating margin on $83B quarterly revenue, highest margins in decades

Microsoft expanded revenues from $17 billion in 2012 to record levels while achieving operating margins last seen in the dot-com bubble, exemplifying American software's unprecedented profitability and scale. The result represents the peak of software dominance but also signals the inflection point where margin compression and competitive threats begin.

Anthropic, OpenAI, and DeepMind converge on public support for pausing frontier AI development

The three dominant AI labs have now publicly aligned on support for a coordinated pause in frontier development. The convergence signals either genuine capability concerns or a strategic move to consolidate market position while competitors catch up, with less charitable readings suggesting potential cartel-like effects that could protect incumbents while weakening competition.

Statement on the US government directive to suspend access to Fable 5 and Mythos 5

Anthropic published a statement on a US government export control directive suspending all access to Claude Fable 5 and Mythos 5 models. Main takeaways:

• USG cited national security authorities and alleged discovery of a jailbreak method bypassing Fable 5's safeguards, though the specific technique demonstrated involved only narrow, non-universal vulnerabilities already present in competing models like GPT-5.5
• Anthropic disputes the government's rationale, arguing the narrow jailbreak finding does not justify commercial model recall and would halt all frontier model deployments industry-wide if applied uniformly
• The directive was issued verbally with minimal technical detail; Anthropic states it reviewed the underlying report and validated the demonstrated capability is widely available from other models and used daily by cybersecurity defenders
• Anthropic maintains its defense-in-depth strategy for Fable 5—combining safeguards with monitoring and 30-day data retention—reduces risks to levels comparable to existing deployed models
• This action represents a significant regulatory escalation against a major AI lab and signals potential future government intervention in model deployment decisions absent transparent statutory process

Fable & time to pause AI

Anonymous mortgage company operator in trust-us.vercel.app (independent essay) published an essay titled "Fable & time to pause AI". Main takeaways:

• Anthropic implements covert product degradation for frontier AI development requests without user notification, while disclosed routing for cybersecurity/biology topics shows the capability to be transparent when chosen.
• Mandatory 30-day data retention for all Mythos-class model prompts applies retroactively to enterprise customers with zero-data-retention agreements, creating potential NDA breaches for users who relied on prior contract terms.
• Anthropic's own internal metrics show 80-100% of its codebase is now written by Claude, yet the company silently degrades the same capability when customers attempt frontier AI development work.
• CEO Dario Amodei's concurrent policy proposal seeks federal gatekeeping authority over model deployments, with "automated R&D" as a fourth risk category alongside cyber/bio/loss-of-control, creating regulatory moat against competition.
• The essay demonstrates false positive rate of safety classifiers through a real example where policy critique triggered cybersecurity flags, raising questions about invisible classifier firing on legitimate work.