[81744] Crisis at Anthropic: US orders shutdown of advanced models Fable 5 and Mythos 5
[81744] Crisis at Anthropic: US orders shutdown of advanced models Fable 5 and Mythos 5
Following the US government's demand for security reasons, Anthropic has disabled access to its leading models for all users and foreign employees.
Drama in the AI industry: Sudden shutdown of Anthropic models
Shake-up in the world of artificial intelligence: Anthropic has announced an immediate shutdown of access to its advanced models, Fable 5 and Mythos 5. The decision follows an export control directive from the United States government, citing "national security" considerations. The order, which enforces a ban on access by foreign nationals (even within the US), effectively resulted in a sweeping block of the models for all users, including paid subscribers and API users, reports the Technology News channel.
According to LΣҒΔ𝕽ΩLL, a channel highlighting the precedent-setting implications of the event, the official background for the directive is the government's claim of identifying a "jailbreak" method in the Fable 5 model. However, Anthropic claims this is a "very narrow case" involving the model's ability to read and fix code, and expresses sharp opposition to the drastic step which, according to them, could "paralyze the deployment of advanced AI models throughout the industry," as reported by Technology and High-Tech News.
Reports indicate that this is a significant policy shift. Even before the shutdown, it was noted by Technology and High-Tech News that Anthropic had attempted to increase transparency regarding the rejection of user requests due to security considerations, but it appears the US government has hardened its stance. News 360 emphasizes that the company described the move as a "misunderstanding," while the company's other models continue to operate as usual.
Experts and commentators in the field, as evidenced by the analysis in LΣҒΔ𝕽ΩLL, warn that this is a "dangerous precedent": according to them, if a commercial model can be taken offline without technical transparency, any AI company could find itself under complete government censorship if someone at the official level decides the model is "too strong" or "too dangerous."