Anthropic suffered a leak of a draft blog post about a new AI model called “Claude Mythos,” which the company is testing with a small group of early access customers and described as “a step change” in performance and “the most capable we’ve built to date.” The incident exposed around 3,000 assets in a publicly accessible data store and coincided with market movements, including bitcoin tumbling from near $70,000 to about $66,000 and notable declines in software stocks such as Palo Alto Networks, CrowdStrike and Fortinet.
A leaked draft blog post referred to a new Anthropic AI model as “Claude Mythos,” and the company described the model in the draft as “a step change” in performance and “the most capable we’ve built to date.” The draft indicated that the model is being tested with a small group of early access customers, and the materials in the leak focused on the model’s capabilities and tiering rather than offering full technical specifications. The leaked content presented high-level characterizations of capability and scale without providing detailed architecture, training data, or evaluation metrics. The draft language emphasized a notable advancement in Anthropic’s product lineup while remaining at a descriptive level in the leaked copy.
Anthropic currently offers three tiers named Opus, Sonnet and Haiku, and the leaked materials mentioned a proposed new tier called “Capybara.” The leak described Capybara as larger and more intelligent than the existing Opus tier, presenting it as an additional tier above the current offerings.
The leaked materials prompted explicit cybersecurity concerns, noting that the model “could pose serious cybersecurity risks, pointing to its ability to identify and exploit software vulnerabilities.” The incident also exposed around 3,000 assets linked to Anthropic’s blog in a publicly accessible data store. The leaked draft emphasized the model’s capabilities and tiering while stopping short of detailed technical specifications. Reporting framed these disclosures as raising immediate questions about potential misuse of advanced capabilities.
Markets reacted during the same period, with bitcoin tumbling to about $66,000 after flirting with $70,000 hours earlier. Notable software-linked names moved lower, with Palo Alto Networks, CrowdStrike and Fortinet reported down roughly 4%–6%, and the iShares Expanded Tech-Software Sector ETF (IGV) down about 2.5%. These price movements were reported alongside the disclosure of the leaked materials. Coverage maintained a cautious, alert tone about the security implications and the possible links between the leak and short-term market volatility.
The event combined an exposure of technical materials with contemporaneous market declines in both cryptocurrencies and software equities. Reporting remained cautious about the security concerns and their implications for financial stability.
The article reports stablecoins evolving into core financial infrastructure, noting North America’s leadership in regulation and institutional adoption. It states that RLUSD surpassed $1 billion in market capitalization within its first year. The reporting lists regulated stablecoin issuers including USDC, RLUSD and PYUSD. The stablecoin items were presented in the article as separate financial developments from other coverage.
Separately, the article reports that the parent company of the New York Stock Exchange is committing nearly $2 billion to prediction markets. The commitment is mentioned among the financial developments covered in the piece. The mention of the NYSE parent’s investment was reported independently from the Anthropic material discussed elsewhere in the article.
These stablecoin developments and the NYSE parent’s prediction-market commitment were presented as distinct financial items in the same reporting. The article did not conflate these financial developments with the Anthropic model leak.
The leak of Anthropic’s draft blog post about Claude Mythos exposed internal materials and prompted concerns about advanced model capabilities. Markets reacted concurrently, with declines in cryptocurrency and software equities reported alongside the disclosures. Coverage also highlighted cybersecurity worries over the model’s potential to find software vulnerabilities and noted separate reporting on stablecoins and a significant investment by the NYSE parent in prediction markets.
