A significant data breach has unveiled Anthropic’s latest AI model, dubbed ‘Capybara,’ which the company claims surpasses all its previous creations in capability. Discovered in a draft blog post left in an unsecured data cache, this leak has raised alarms about cybersecurity implications.
The revelation, reported by Fortune, indicates that Anthropic’s new model, referred to internally as ‘Mythos,’ is described as the most powerful AI model the company has developed to date. Cybersecurity researchers found this information among nearly 3,000 unpublished assets in a publicly accessible database.
Following inquiries from Fortune, Anthropic confirmed the model’s existence, emphasizing its potential as a ‘step change’ in AI performance. The company acknowledged that a ‘human error’ in its content management system led to the data exposure, which has since been rectified.
“Compared to our previous best model, Claude Opus 4.6, Capybara gets dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity, among others,” the draft stated.
Of particular concern is the model’s implications for cybersecurity, especially within the cryptocurrency sector. The draft warned that Capybara poses ‘unprecedented cybersecurity risks,’ which could impact blockchain security and smart contract auditing, amidst an ongoing arms race between cyber attackers and defenders.
This week, Ripple announced an AI-driven security enhancement for its XRP Ledger after an AI-assisted team identified over ten vulnerabilities in its 13-year-old code. Concurrently, Ethereum launched a post-quantum security hub, reflecting the urgent need for advanced security measures.
In a separate incident, the Resolv stablecoin lost its peg after an attacker exploited vulnerabilities in a minting contract. Such incidents highlight the necessity for more sophisticated AI tools to detect and mitigate risks before they can be exploited.
The leak also raises competitive questions in the AI token market. Recently, Bittensor introduced a model named Covenant-72B, competing with Meta’s Llama 2 70B, which resulted in a significant rally for the TAO token and increased market capitalization for subnet tokens.
Anthropic has indicated it will proceed cautiously with the release of Capybara, citing its high operational costs and the current unavailability for general use. The company has since restricted public access to the compromised data cache.
This incident serves as a stark reminder of the potential vulnerabilities even in organizations developing advanced cybersecurity solutions, highlighting the irony of a major AI model announcement being mishandled.
A data leak revealed Anthropic's new AI model 'Capybara,' which is claimed to be the most powerful to date. The breach raises significant cybersecurity concerns, especially for the cryptocurrency sector, as the company acknowledges the risks associated with its advanced capabilities.
Source: Fortune