Anthropic: Fear of China espionage allegedly triggers AI lockdown

Details suggest: Because Korean partner SK Telecom allegedly has ties to Beijing, the White House withdrew its trust in Anthropic for Mythos.

listen Print view

(Image: Anthropic / Bearbeitung heise medien)

4 min. read

The Trump administration is taking a hard line in the global AI race. The trigger for far-reaching export controls against the AI company Anthropic is a simmering conflict over access to its most powerful technology, the Claude Mythos model. As part of its Project Glasswing program, Anthropic also made this system available to the South Korean telecommunications giant SK Telecom, according to Wired. This reportedly alerted the White House security authorities, who suspected covert ties to China behind the South Korean company.

Nervousness in Washington regarding a potential Beijing connection intensified when Amazon alerted the White House to vulnerabilities in Fable 5. This is a modified version of Mythos that Anthropic released to the public in early June. Amazon researchers demonstrated that the built-in safeguards could be bypassed, allowing unauthorized users to gain access to Mythos's cybersecurity capabilities.

For the US government, the pieces of the puzzle came together to form a threat scenario: A powerful AI tool capable of detecting software security vulnerabilities was apparently in the hands of a partner with ties to China. Added to this was the fear of the potential activation of all functions.

Videos by heise

Through export restrictions, the White House ordered the withdrawal of access to Mythos and Fable 5 for all foreign nationals to cut off any data flow to Beijing.

Since filtering by nationality is technically difficult to implement in a data protection-compliant manner, Anthropic pulled the emergency brake and shut down the affected models. The fact that the US government reacted so uncompromisingly in this case is likely also due to Amodei's refusal to submit politically. The Pentagon had already shown Anthropic the door in March due to a dispute over mass surveillance.

The fact that SK Telecom, in particular, came under the scrutiny of the US hawks is historically justified. Although the mobile communications division recently generated hardly any revenue in China and denies any improper proximity to Beijing, SK Telecom is part of a larger conglomerate whose subsidiaries conduct semiconductor and energy business with the People's Republic.

Furthermore, SK Telecom has maintained the joint venture UNISK with the state-owned Chinese telecommunications provider China Unicom for over twenty years. The Trump administration banned this company early on due to alleged ties to the Chinese military with investment bans. The US regulator FCC banned it from the US network. For the White House, this was reason enough to pull the plug.

Meanwhile, Anthropic is trying to limit the damage in Washington. The startup has tripled its lobbying presence on Capitol Hill this year, opened an office in Washington, and hired high-profile Republicans and former Trump staff. Both sides are currently negotiating daily on a solution, where a joint technical review process could pave the way for a restart. In the tech industry, concerns are growing that the China suspicion is setting a precedent and that an "electronic citizenship check" will become a standard requirement in Silicon Valley.

Critics accuse Amodei of underestimating the geopolitical implications and the China paranoia in Washington. His warnings have become his undoing: Amodei had previously emphasized that Mythos was too powerful for sophisticated cyberattacks to be released without control. Although over 175 tech CEOs and cybersecurity experts rallied to Anthropic's side in an open letter after the forced shutdown, the trust of the US security authorities seems to have been lost.

In contrast, OpenAI CEO Sam Altman immediately adopted a conciliatory approach after Trump's election victory. The ChatGPT operator hired former presidential staff, financed US data centers in line with the administration's wishes, and sent money through lobbying groups to politicians who oppose AI regulation. The reward for political adaptation was evident at the recent G7 summit: Altman sat directly next to Trump at the AI lunch. Amodei had to make do with French President Emmanuel Macron.

(des)

Don't miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.