AI Security
Anthropic breached in hours by AI startup during bug bounty test
AI firm Anthropic had its internal network breached in under 12 hours by startup Cathedral during a bug bounty programme. Exposed data included staff salaries and internal messages, but user chats and core code were not accessed.
Ross Grant2 min read
The short version
Anthropic, a well-known AI company, was hacked by another startup called Cathedral during a bug bounty test. The attackers got into Anthropic's internal systems and accessed staff messages and salary details. However, they did not get any user chats or the main AI code. The company stopped the attackers quickly and confirmed that customer data was not affected.
By the numbers
- 12 hourstime taken for Cathedral to breach Anthropic's networkThe Independent
- 0user conversations and core AI code compromisedThe Independent
What happened?
AI startup Cathedral breached Anthropic's internal network in under 12 hours during a bug bounty programme, according to The Independent. The attackers accessed internal Slack messages, staff salary details, and meeting transcripts. Anthropic said user conversations and core AI source code were not compromised. Cathedral's access was revoked the same day.
Who is affected
Anthropic, a major US artificial intelligence company, was breached by another AI startup called Cathedral. The incident happened during a bug bounty programme, where companies invite others to find security flaws in exchange for rewards. The breach exposed internal data belonging to Anthropic staff, but not its customers.
What was taken
According to The Independent (opens in a new tab), Cathedral accessed internal Slack messages, staff salary information, and meeting transcripts. Anthropic confirmed that no user conversations with its Claude chatbot were compromised. The company also said its core AI source code was not accessed.
How the attack worked
Security experts told The Independent that the breach was due to a basic supply-chain hygiene failure. This means the attackers took advantage of weaknesses in third-party software used by Anthropic, rather than directly attacking the company's main AI systems. It was not a sophisticated attack on the AI infrastructure itself. Cathedral managed to gain access to Anthropic’s internal network in less than 12 hours after starting the bug bounty challenge. Anthropic cut off Cathedral’s access the same day.
The response
Anthropic responded quickly by revoking Cathedral’s access as soon as the breach was discovered. The company confirmed that sensitive user data and its core AI code were safe. The incident was made public shortly after Anthropic filed for an initial public offering (IPO), which is expected to be the largest stock listing in history. An anonymous former Trump administration official told The Independent that the breach shows vulnerabilities that could be exploited by foreign adversaries, such as China. However, there is no evidence that any nation-state was involved in this case.
Why it matters
This incident shows that even leading AI companies can be vulnerable to attacks through third-party software. Sensitive internal information, like staff salaries and internal discussions, can be valuable targets. The breach highlights the importance of supply-chain security, especially as AI firms grow and prepare for major events like IPOs.
Key points
- Anthropic's internal network was breached in under 12 hours by Cathedral.
- Exposed data included staff Slack messages, salaries, and meeting transcripts.
- User conversations and core AI source code were not compromised.
- The breach was due to a third-party software weakness, not a direct attack on AI systems.
- Anthropic revoked access quickly and confirmed the scope of the breach.
Sources
The Independent
Reference
