This week, the artificial intelligence firm Anthropic announced the development of a highly advanced AI model, named Mythos, which it has chosen not to make publicly available due to a sense of profound responsibility.
Scott Bessent, the U.S. Treasury Secretary, convened leaders from major banking institutions to discuss Mythos. In a letter to the government, Danny Kruger, a Member of Parliament with Reform UK, urged officials to collaborate with Anthropic, indicating that the new model could pose severe cybersecurity threats to the United Kingdom. Social media platforms erupted in reaction.
However, skepticism emerged from some quarters, particularly from prominent AI critic Gary Marcus. He remarked that while Dario Amodei, Anthropic’s CEO, possesses more technical expertise than Sam Altman of OpenAI, he appears to engage in similar levels of hype and exaggeration.
While it remains uncertain if Anthropic has indeed created a transformative AI, it is clear that the San Francisco-based startup, regarded as a “responsible” AI entity, has excelled in its marketing efforts.
In recent months, Anthropic has garnered significant media attention, including a lengthy feature in The New Yorker, articles in The Wall Street Journal, and a cover story in Time magazine, where Amodei’s image was prominently displayed alongside the Pentagon and U.S. Defense Secretary Pete Hegseth.
In February, Amodei and co-founder Jack Clark participated in two different podcasts with The New York Times, discussing topics like the consciousness of their AI and its potential economic impact. The company’s “resident philosopher” has also engaged with The Wall Street Journal, pondering whether Claude, a product utilized for cryptocurrency trading and military targeting, possesses a “sense of self.”
This media flurry coincided with a contentious situation between Anthropic and the U.S. Department of Defense. Despite having developed the AI tool employed by the Pentagon for operations in Iran, Anthropic has managed to maintain a more favorable public image than OpenAI, which offered military assistance but with less oversight.
Danielle Ghiglieri, Anthropic’s media lead, celebrated these media successes on LinkedIn. She expressed pride in working for Anthropic and highlighted the rush to finalize the Time magazine story as a significant achievement.
Reflecting on the New Yorker profile by journalist Gideon Lewis-Kraus, she admitted to feeling nervous about their first in-person meeting, noting that collaborating with such a high-caliber journalist pushed her to refine her ideas.
Other public relations professionals in the tech industry have taken notice of Anthropic’s current prominence. One remarked that while the company is currently in the spotlight, any firm developing world-changing technology deserves rigorous scrutiny.
Recently, Anthropic inadvertently disclosed part of Claude’s internal source code, although it assured that no sensitive customer information was compromised.
The implications of Anthropic’s powerful model, Mythos, remain under examination. Dr. Heidy Khlaaf, chief AI scientist at the AI Now Institute, indicated that the claims surrounding Mythos had not been adequately verified. She suggested that vague marketing language could indicate an attempt to attract further investment without proper oversight.
Jameison O’Reilly, an offensive cybersecurity expert, acknowledged Mythos as a legitimate advancement but pointed out that some of Anthropic’s assertions, such as discovering thousands of “zero-day vulnerabilities” in major operating systems, may not significantly impact real-world cybersecurity.
O’Reilly explained that in over a decade of gaining authorized access to various organizations, the instances requiring a zero-day vulnerability were exceedingly rare.
Other factors may have influenced Anthropic’s choice not to release Mythos. The company faces resource constraints and seems to be struggling to provide sufficient computing power for all its users. Recently, it implemented usage limits on the popular Claude model and indicated that users would need to buy additional capacity to utilize third-party tools.
Like OpenAI, Anthropic is in a race to secure substantial funding and capture a still-evolving market for AI-driven chatbots that could serve as companions, romantic interests, or highly personalized assistants, as well as for businesses looking to substitute human labor.
However, the differences between their products are minimal and subjective, often hinging on intangible qualities attributed to AI, such as a “sense of self” or “soul.” The competition is as much about public perception as it is about technology.
Dr. Khlaaf described the announcement of Mythos as a strategic move to signal that Anthropic is ready to engage in the market, although the limitations on its release restrict independent assessment of the company’s claims.
She suggested that this might reflect a familiar pattern of marketing tactics seen in the tech industry.
