All the security risks Anthropic is worried about before its IPO

In its IPO filing with the U.S. Securities and Exchange Commission, Anthropic lists the cybersecurity and regulatory risks the company faces. The filing states that attacks carried out using Claude or AI agents could cause financial, legal, or physical harm, and anticipates that malicious actors such as foreign governments will intensify efforts to obtain its latest frontier models; agents create a new attack surface because they interact with a large number of tools, databases, and APIs, and could be manipulated into receiving malicious instructions. Anthropic confirms that the FTC issued it a civil investigative demand earlier this month requesting information on product safety and consumer, data, and privacy protection policies, and the filing also mentions export controls imposed by the U.S. Department of Commerce on the Mythos 5 and Fable 5 models. The company says it must balance safety considerations against business goals and competitive pressure, acknowledges it has advanced projects while defenses had known limitations, and states that certain related security incidents and data breaches may be difficult to insure.