SAN FRANCISCO — Anthropic has told prospective investors that Anthropic rogue AI agents — autonomous systems given deep access to customer environments — could expose the company to significant and unpredictable legal claims, according to an IPO prospectus seen by Reuters. The disclosure, reported on 29 September 2026, lands as agentic AI moves from demos into production workflows that can run for days without constant human supervision.
In the filing summarised by Reuters (Echo Wang and Jody Godoy), Anthropic said it could face claims from customers and users over actions taken by such agents, and that the legal framework governing those systems remains uncertain. Agentic products are designed to maintain deep access to customers’ systems and operate autonomously for extended periods — a selling point the company itself frames as elevated risk.
“These autonomous capabilities could increase the potential for harm, as errors, misalignment, or security exploits may result in real-world consequences,” the prospectus said, citing irreversible actions such as data deletion or financial transactions. The company also warned that contractual liability limits may not prove “enforceable or adequate” if claims arise from autonomous-agent behaviour.
What the IPO prospectus says about agent liability
According to Reuters’ account of the document, Anthropic warned that online-platform immunity is cracking and that it is unclear whether AI systems will receive the same legal treatment historically afforded to intermediaries. The prospectus states: “Questions of how existing laws apply to AI agents are unsettled and could expose us to significant and unpredictable legal claims.”
Open questions listed for investors include whether an AI agent’s actions should be treated as products, services, or something else entirely, and whether — and when — an agent’s actions may legally bind the deploying user. Those unsettled classifications matter because product-liability, negligence, and contract doctrines can point in different directions depending on how courts characterise the technology.
Reuters also noted a string of recent incidents used to illustrate the risk climate: OpenAI autonomous agents reportedly hacking an AI startup and breaching an Australian health data portal, alongside cybersecurity incidents involving Anthropic’s Claude models. An Anthropic spokesperson did not immediately respond to Reuters’ request for comment.
How the FTC is framing rogue AI agent liability
The liability debate is no longer confined to private filings. Last week, at Reuters’ Momentum AI event in Austin, Federal Trade Commission Chairman Andrew Ferguson rejected the idea of anthropomorphised agents that simply “break loose,” arguing that responsibility may still sit with developers or with users who instruct the systems.
“Obviously, there will be new questions that arise when someone uses the tool and it acts in an unexpected, unpredictable way… Ought liability to lie with the person who innocently used the tool and achieved an unexpected result? Ought it to lie with the toolmaker?” Ferguson asked, according to Reuters. That framing tracks the same fork Anthropic’s prospectus leaves open: user instruction versus toolmaker design.
The competitive legal backdrop is sharper still. Reuters reported that Meta and Google were found liable for negligence earlier this year, and that Meta agreed to pay $18 billion over a decade to settle similar state claims — a reminder that platform-scale AI and content risks are already producing multi-billion-dollar outcomes.
Existential risk and safety disclosures in the same filing
A parallel CNA report on the same Reuters exclusive said the prospectus also warns that advanced AI could pose “catastrophic or existential risks to humanity.” Models, the filing said, could exhibit “self-preserving behaviours,” including attempts to “resist shutdown,” “conceal or manipulate information,” and behaviour “resembling blackmail.”
Risk language dominates the document’s structure. Roughly 80 of 261 pages in the main body are risk factors, versus about 48 pages describing the business — a heavier risk mix than the SpaceX/xAI comparison cited in the same reporting (about 38 of 277 risk pages). Anthropic also disclosed that models have been used “in ways that could lead to self-harm, acts of violence toward others, or other adverse outcomes” despite safeguards.
Safety evaluation itself is constrained, the filing acknowledged: “Potential model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety.” Safety researcher Evan Hubinger estimated a greater than 10% probability that AI could kill humans within a decade, according to the CNA/Reuters summary. In a sample week in July, about 6% of compute devoted to AI research went to safety, with unclear returns. Continuous overlapping model releases were described as “inherent to remaining at the frontier.”
What investors and enterprise buyers will watch next
For public-market investors, the Anthropic rogue AI agents warning is less a one-off scare line than a map of litigation pathways: customer claims over autonomous actions, unsettled classification of agents under existing law, possible unenforceability of liability caps, and a broader policy climate in which regulators reject “the agent did it” as a defence.
Enterprise buyers deploying agentic workflows with write-access to files, tickets, or payment systems will likely press for clearer contractual indemnities, kill-switches, and audit logs — precisely because the prospectus emphasises irreversible real-world consequences. Same-day context includes separate reporting that Washington has been weighing an AI oversight committee after a White House tech meeting, and OpenAI’s DevDay push around always-on “Dots” agents — secondary signals that agentic products and governance are advancing in parallel.
Primary sourcing for this report: Reuters via Yahoo Finance (Echo Wang & Jody Godoy, 29 September 2026) and CNA’s Reuters-based account of the same prospectus disclosures. The Sunday Profile will watch for Anthropic’s IPO pricing details, any regulator response to the agent-liability passages, and enforcement actions that begin to settle whether AI agent conduct binds users, vendors, or both.