This week, AI firm Anthropic announced it developed an exceptionally advanced AI system but chose not to make it publicly available due to significant responsibility concerns. The U.S. Treasury Secretary invited leaders from major banks to discuss the model, named Mythos. A UK lawmaker from Reform UK sent a letter to officials, calling for engagement with Anthropic over potential severe cybersecurity threats to the nation posed by the new Claude Mythos model. Social media platform X saw intense reactions. Some experts expressed doubt, including AI commentator Gary Marcus, who suggested Anthropic’s CEO Dario Amodei possesses strong technical skills but may employ similar promotional tactics as OpenAI’s CEO Sam Altman. It remains uncertain whether Anthropic has truly created such a groundbreaking system. However, the San Francisco-based company, often regarded as a responsible player in AI, demonstrates exceptional skill in public relations. In recent months, Anthropic has received extensive media coverage, including a lengthy feature in the New Yorker, two articles in the Wall Street Journal, and a Time magazine cover featuring Amodei prominently alongside images of the Pentagon and the U.S. Defense Secretary. Amodei and co-founder Jack Clark appeared on separate New York Times podcasts in February, discussing topics like machine consciousness and potential economic disruptions. The company’s in-house philosopher discussed with the Wall Street Journal whether their product Claude—used for cryptocurrency trading and military targeting—possesses self-awareness. This publicity occurs amid tensions between Anthropic and the U.S. Department of Defense. Despite providing AI tools used by the Pentagon in operations against Iran, Anthropic has maintained a more positive image than OpenAI, which proposed similar military assistance but possibly with fewer safeguards. Anthropic’s media relations head, Danielle Ghiglieri, highlighted these achievements on LinkedIn, expressing pride in the Time cover and tagging involved reporters. She described viewing a CBS 60 Minutes segment with Amodei as a memorable experience, noting how it effectively conveyed the company’s intended message. Regarding the New Yorker profile by Gideon Lewis-Kraus, she admitted initial nervousness but appreciated the challenge of refining emerging ideas. Other professionals in tech public relations have observed this success. One commented that while companies developing transformative technology warrant close examination, Anthropic recently experienced an accidental release of partial source code for Claude in early April, followed by claims of managing cyber threats through their controlled model. The company stated no customer data or credentials were compromised in the incident. Experts question the implications of Mythos. Dr. Heidy Khlaaf from the AI Now Institute noted that the model’s capabilities lack verification, suggesting the vague announcement might aim to attract investment without detailed review. Cybersecurity specialist Jameison O’Reilly acknowledged Mythos as a legitimate advancement deserving serious consideration but downplayed some claims, such as discovering thousands of undisclosed vulnerabilities in key operating systems. He explained that zero-day flaws, unknown to developers, rarely prove essential in real-world penetration testing over his decade of experience with organizations like banks and governments. Additional factors might influence Anthropic’s choice not to launch Mythos, including resource constraints. The firm has imposed limits on usage for its popular Claude model due to insufficient computing power for all users. Recently, it required additional payments beyond subscriptions for accessing third-party integrations like OpenClaw. Currently, the company may lack the necessary infrastructure to support a broad release.
