Anthropic Outages Stoke Security Fears at Coinbase

Anthropic experienced service disruptions on Wednesday that affected its Claude chatbot, API and developer tools, including the Mythos model. The outages prompted security warnings from Coinbase and other firms that manage exchange, custody and wallet software.

Anthropic reported elevated error rates beginning at 10:53 a.m. ET. The company said login success rates for the chatbot stabilized around 12:30 p.m. ET and that all systems were restored by 1:50 p.m. ET. Outage monitoring services recorded roughly 6,000 user reports at the height of the incident; the number of reports fell sharply as services recovered.

Philip Martin, Coinbase’s chief security officer, warned that Mythos and similar models ‘will enable even deeper testing of software and systems at scale.’ He added such models ‘will accelerate digital threats as well as digital defense,’ and described Mythos as ‘a highly restricted model not available to the public.’ Martin said Coinbase is in close contact with Anthropic about potential security implications.

The concern from exchanges and custodians centers on the software and infrastructure that hold private keys and run trading systems, rather than the Bitcoin ledger. The Bitcoin blockchain has operated continuously since 2009 and has not been breached.

Access to advanced models has not been uniform. Some large banks and technology firms received early access to Mythos, while many crypto companies were not part of the initial rollout. Binance’s chief security officer, Jimmy Su, said the exchange relies on a mix of public large language models, earlier Anthropic systems and internal tools to address security needs.

Trading and market commentary reacted to the outage and security discussion. Coinbase shares rose more than 7% to about $198 during trading. Piper Sandler raised its price target for Coinbase to $180 from $150 while keeping a Neutral rating, citing difficult year-over-year comparisons for exchanges in the second quarter.

Anthropic, founded in 2021 by researchers and executives who left OpenAI, has expanded quickly and attracted major investment; those funding rounds have pushed the company’s valuation to about $800 billion. OpenAI’s most recent funding round valued it near $850 billion.

Security teams at exchanges, custodians and wallet providers are preparing for faster, more automated probes of code and systems as AI models improve. Firms say they plan to increase testing and defensive measures in response to the changing capabilities of such models.