Powell and Bessent unite against risks from Anthropic’s AI models

---

---

Supposed rivals Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell pulled the heads of the biggest U.S. banks into an urgent meeting in Washington over concerns inside government circles over what Anthropic’s newest AI system could mean for cyber risk.

According to claims from Bloomberg, the meeting allegedly happened on Tuesday at Treasury headquarters and discussed locking things down before tools like Anthropic’s Mythos get used against critical financial systems.

The executives called in were running banks that regulators treat as essential to the financial system. Those invited included Jane Fraser of Citigroup, Ted Pick of Morgan Stanley, Brian Moynihan of Bank of America, Charlie Scharf of Wells Fargo, and David Solomon of Goldman Sachs. Jamie Dimon of JPMorgan was unable to attend.

Crypto community’s Arthur Hayes reacted on X with a jab at the moment, writing, “Powell and Bessent provided the cancerous soft drink to go with the Trump taco.” He added a screenshot of the USD liquidity index covering a one-year period.

Source: Arthur Hayes/X

Moving on, Scott and Powell reportedly wanted the banks to understand the possible danger tied to Anthropic’s Mythos and to similar models that may come after it. They also wanted the companies to take steps now to protect their systems, because it seems that’s what matters most, and not the general public.

Earlier, Cryptopolitan reported that Anthropic has said Mythos can find weak points in every major operating system and web browser and then use those weak points when a user tells it to do so.

Anthropic chases more chips while OpenAI readies a cyber product for select partners

Meanwhile, Reuters reported earlier that Anthropic is looking at whether it should design its own chips. Three sources said the company is exploring that option as AI companies deal with a shortage of the chips needed to train and run more advanced models.

Those plans are still early. Two people with knowledge of the matter and one person briefed on Anthropic’s plans said the company could still decide not to build its own chips at all and just keep buying them.

The backdrop is a surge in demand for Claude. Anthropic said earlier this week that its run-rate revenue is now above $30 billion in 2026, up from about $9 billion at the end of 2025.

To build and run Claude, Anthropic uses several kinds of chips, including TPUs made by Google and chips from Amazon.

Earlier this week, Anthropic also signed a long-term deal with Google and Broadcom, which helps design those TPUs. That deal adds to the company’s commitment to spend $50 billion on stronger U.S. computing infrastructure.

Anthropic’s rivals, Meta and OpenAI, are both pursuing their own chip efforts as well. Industry sources said building an advanced AI chip can cost around $500 million because companies need top engineers and heavy spending to reduce manufacturing defects.

At the same time, OpenAI is working on its own cyber product. A source familiar told Axios that OpenAI is finishing a product with advanced cybersecurity features and plans to release it to a small group of partners. Back in February, after launching GPT-5.3-Codex, OpenAI introduced its Trusted Access for Cyber pilot. The company said in a blog post that groups in the invite-only program would get access to “even more cyber capable or permissive models to accelerate legitimate defensive work.” At that time, OpenAI also set aside $10 million in API credits for participants.

Your keys, your card. Spend without giving up custody and earn 8%+ yield on your balance with Ether.fi Cash.