What Anthropic’s Mythos-class Fable 5 means for CEOs governing AI: ‘Oh God, no! Not another thing:’



Good morning. I was speaking with a CEO in financial services this week when I noticed and mentioned that Anthropic had just released its first Mythos-class model to the public. His reaction: “Oh God, no! Not another thing.” Neither of us knew the details yet, beyond the now disputed claim that Claude Fable 5 is a better way to let users query a model deemed too dangerous for general release. To him, that didn’t matter. “I’m not worried about someone making a weapon,” he told me. “I’m worried about all the other crap we now have to think about.”

That reaction captures something every CEO needs to grapple with as the rules of the road on AI keep changing and the companies writing them don’t seem to answer to anyone but themselves. Anthropic’s rollout of Claude Fable 5 is a case study in the new reality of AI adoption, creating new challenges along with new capabilities. 

To start, there’s the trade-off of having Fable 5 automatically block responses in areas like cybersecurity and even limit the capabilities of AI researchers and developers. (The unfettered version—Claude Mythos 5—is available only to pre-approved organizations to address cybersecurity issues.) Another problem is that the model holds back covertly: no red flag or transparency about triggers, so users don’t know when they’re being served up subpar content. They just know from the system card that it could happen any time.

Then there’s the data problem. Anthropic is a favorite with health care leaders like Banner Health CEO Amy Perry who spoke with me recently about their digital transformation and moves to modernize while protecting 29 petabytes of data. “We felt the Anthropic model was a safer platform that allows us to apply the kinds of governance that we wanted to apply,” she told me. It was also HIPAA-compliant in offering zero data retention. Claude Fable 5 does not, instead maintaining a 30-day data retention window that has prompted companies like Microsoft to reportedly restrict employees from using it.

And of course, there are the cybersecurity concerns. On that front, I had a wide-ranging conversation yesterday with Bezalel Eithan Raviv, CEO of Tel Aviv-based Lionsgate Intelligence Network. Like a lot of cybersecurity experts I speak with, Raviv takes issue with the lack of guardrails around these products. “Anthropic officially has become the judge and the executioner in a way, they decide the standard for what will be right or wrong, even though it’s a standard that nobody agreed to.”

In his view, these models are more akin to currency or weapons than products and should be treated as such. “You can’t print U.S. dollars or produce weapons because you feel like it,” he said. “This area demands a shepherd. We’re seeing a lot of new sheep. Some are like (the cloned) Dolly where it’s hard to understand the origin or implications of what you’re seeing. Cybercriminals are using these weapons to intercept human beings through sophisticated attacks … Anthropic is a private company, with shareholders who want to maximize profit. They’ve reached the point where even they don’t know how to anticipate what’s next. It’s madness.”

A final, unrelated note: Be sure to check out the inaugural Fortune Crypto 100 that’s out today. Our editorial team has ranked the most influential companies and protocols in the digital asset ecosystem—from crypto-native pioneers like Coinbase and Uniswap to Wall Street giants like JPMorgan and BlackRock. See who made the list, and who came out on top in all 10 categories here.

Contact CEO Daily via Diane Brady at diane.brady@fortune.com

Top leadership news

H-1B visa fee chaos

A federal judge struck down Trump’s $100,000 H-1B visa fee this week, but AI firms like OpenAI and Nvidia had already been forking over money for the tax. The case now appears headed for the Supreme Court, with conflicting rulings leaving employers uncertain whether the fee still applies today.

Meta woos data center workers

Meta is launching a free five-week training program for data center technicians, covering tuition, housing, and a daily stipend, as the company races to staff its $600 billion U.S. data center buildout. “America needs hundreds of thousands of skilled tradespeople,” said Meta VP of data centers Rachel Peterson.

4.2% inflation

Inflation jumped to 4.2% in May, driven largely by a 40% annual surge in gasoline prices tied to the Iran war, putting new Fed chair Kevin Warsh in a bind just weeks into his tenure. Still, some analysts see a silver lining: strip out energy, and core inflation actually cooled, suggesting the spike may be temporary.

The markets

S&P 500 futures are up 0.79% this morning. The last session closed down 1.62%. The STOXX Europe 600 was up 0.49% in early trading. The U.K.’s FTSE 100 was up 0.59% in early trading. Japan’s Nikkei 225 was up 0.06%. South Korea’s KOSPI is up 0.43% today. China’s CSI 300 was down 0.55%. Hong Kong’s Hang Seng was down 0.65%. India’s NIFTY 50 was flat. Bitcoin was up at $63K.

Around the watercooler

Analysts expected oil to surge above $200 but China has quietly kept prices half of that—and can’t for much longer by Sasha Rogelberg

Anthropic accused of ‘secret sabotage’ as Claude Fable 5 silently limits capabilities for AI researchers and developers by Sasha Rogelberg

Xbox’s CEO spent her early career taking out trash and selling coupon books—she says the secret to her rise was never obsessing over a dream career by Preston Fore

‘The circulatory system is not working.’ Goldman on what’s really wrong with private markets by Nick Lichtenberg

Exclusive: Mastercard launches protocol to let AI agents pay each other, send micropayments by Ben Weiss

CEO Daily is curated and edited by Joseph Abrams, Jason Ma, Claire Zillman, and Lee Clifford.



Source link