Anthropic's Mythos Warning Was Both Responsible - and Clever Marketing

As today's SpaceX IPO reveals, marketing and scarcity remain among Silicon Valley's most potent assets, something their dotcom predecessors honed to a fine skill. 

While it would be inaccurate to claim the maneuvering around Anthropic's Mythos is misleading, it is fair to say that promotion is still an important part of the tech oeuvre, especially in the early stages before customer experience and knowledge surpasses suppliers optimistic claims. JL

Ashley Capoot reports in CNBC and Blake Montgomery reports in The Guardian :

Two months after Anthropic rolled out Mythos to a limited number of users, citing concerns about the AI model’s potential to do damage, the company is ready to release it to the public. With the launch of Claude Fable 5, Anthropic is honoring its “eventual goal” to deploy Mythos-class models at scale. It’s also capitalizing on growing momentum and investor interest ahead of a massive IPO. Anthropic will be under pressure to justify it’s valuation, and Claude Fable 5 could become a valuable new money-maker. The model costs $10 per million input tokens and $50 per million output tokens, The technology is powerful, but so is the manipulation of attention. “Mythos is a real development and Anthropic was right to treat it seriously,” (but) it is clever marketing to say: “You can’t have this; it’s too strong.” Nothing sparks a stronger desire than withholding. 

Two months after Anthropic rolled out Mythos to a limited number of users, citing concerns about the artificial intelligence model’s potential to do damage in the wrong hands, the company said it’s ready to release an equally powerful model to the public.

Anthropic on Tuesday announced Claude Fable 5, a Mythos-class model that will be available to its enterprise customers and paid subscribers. The company said the broad release is possible because of new safeguards that block responses in specific high-risk areas, including cybersecurity and biology. 

“For us, it’s really around what we call ‘race to the top,’ being able to provide this technology in a valuable fashion, and at the same time providing the right safety guardrails so that it can do asymmetrically more benefits than harm,” Dianne Penn, Anthropic’s head of product management for research, told CNBC in an interview. 

Anthropic captivated Wall Street and government officials in April with the unveiling of Mythos, which excels at identifying security flaws within software. The company said it did not plan to make the model generally available, and it has limited the rollout to a select group of companies as part of a cybersecurity initiative called Project Glasswing.

But with the launch of Claude Fable 5, Anthropic is honoring its stated “eventual goal” to deploy Mythos-class models at scale. It’s also capitalizing on growing momentum and investor interest in its technology ahead of a potentially massive IPO, which is expected to take place as soon as this year. 

Artificial intelligence companies make powerful products. They also make outlandish claims.

When Anthropic released Claude Mythos, an AI model focused on cybersecurity, which has inspired widespread thrill and panic over how capable it is said to be. There was a catch, though: the public can’t touch it. OpenAI declared at the end of last week that it, too, had developed a frighteningly capable cybersecurity AI.

Anthropic called Mythos a “reckoning” for the cybersecurity industry because of how keenly adept at exposing software weaknesses the new tool had proven. According to Anthropic, Mythos has laid bare thousands of vulnerabilities in commonly used applications for which no patch or fix exists, prompting the startup to form an alliance with cybersecurity specialists, nicknamed Project Glasswing, to bolster defenses against hacking and withhold the model from wide distribution, much like an arthouse film that plays solely in Los Angeles and New York. 

Shakeel Hashim writes in a Guardian op-ed:

Mythos, the company claims, has found vulnerabilities in every major browser and operating system. In other words, this new AI model might be able to help hackers disrupt much of the world’s most important software.

If such technology was widely available and as capable as Anthropic claims, the implications could be catastrophic. Cyber-attacks are no longer a solely digital problem. Almost everything we rely on in the physical world involves software. In recent years, airports, hospitals and transport networks have been crippled by cyber-attacks. Until now, attacks of this scale required serious expertise. Mythos would put that capability in reach of amateurs – and turbocharge the professionals’ ability to wreak havoc.

Already, though, cybersecurity experts are pushing back on Anthropic’s claims. My colleague Aisha Down reports:

It is unclear if Anthropic has built the machine god. What is more apparent is that the San Francisco startup widely seen as the “responsible” AI company is brilliant at marketing.

“Mythos is a real development and Anthropic was right to treat it seriously,” said Jameison O’Reilly, an expert in offensive cybersecurity. But, he said, some of Anthropic’s claims, such as that it found thousands of “zero-day vulnerabilities” in major operating systems, were not that significant to real-world cybersecurity considerations.

In 2017, BuzzFeed News’s tech editor argued that the engineering of desire was Apple’s greatest strength. I agree, and Anthropic seems to possess a similar genius. The technology is powerful, but so is the manipulation of attention. Like Apple’s genuinely transformative iPhone, Claude is the real deal. Businesses can’t get enough of its coding capabilities. Serious companies – Apple, Nvidia, Google, JPMorganChase, Amazon Web Services, Broadcom – have partnered with Anthropic on Project Glasswing. However, it is also a clever bit of marketing to say: “You can’t have this; it’s too strong.” Nothing sparks a stronger desire than withholding. According to Bloomberg, Anthropic was the talk of the town at the HumanX AI conference in San Francisco last week.

Hype has obscured the public’s understanding of generative AI since the technology’s advent. Journalists and clear-eyed observers have been trying to pierce it for just as long. In 2019, Slate published a piece, “OpenAI says its text-generating algorithm GPT-2 is too dangerous to release” (hat tip to X’s @Banteg for surfacing the story in a tweet). The headlines may look similar because Anthropic’s CEO, Dario Amodei, was once the vice-president of research at OpenAI. He left the company in 2020. OpenAI also withheld its video generator Sora from wide release for many months. Did that kill Hollywood and film-making as we know it? No. The company shuttered Sora late last month.

Concerns over the hazards of simple text generation might seem quaint in comparison to the complete end of digital security, but the fact that we have moved past the fears of 2019 points to a heartening likelihood: we will move on from today’s fears of cybersecurity’s demise as well, settling in a middle-ground reality between the present status quo and the hyperbolic future.   

Fierce competition means Anthropic CEO Dario Amodei will be under pressure to justify the startup’s valuation to investors, and Claude Fable 5 could become a valuable new money-maker for the company. The model costs $10 per million input tokens and $50 per million output tokens, which is twice as expensive as Claude Opus 4.8.