When Claude's outputs control hardware capable of physical harm (including autonomous vehicles, surgical devices, industrial machinery, and hazardous materials systems), a qualified human observer must maintain real-time ability to halt operation, and all physical safety limits must be enforced by hardware or software independent of the AI model's output.
This analysis describes what Anthropic's agreement states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability and practical outcomes may vary by jurisdiction, enforcement context, and individual circumstances. Read our methodology
This provision establishes hardware-level architectural requirements for AI-physical system integrations, mandating that safety limits be enforced outside the model control loop and that fail-safe behavior occur upon loss of connection to Anthropic's services, which has direct implications for robotics, medical device, industrial automation, and autonomous vehicle deployments.
The updated policy clarifies that Anthropic's acceptable use rules apply to all users of its products and services, including API developers, cloud customers, and app users. The policy now explicitly prohibits weapons development and related activities, and introduces a Cyber Verification Program allowing researchers to apply for adjusted safeguards for authorized dual-use work. Enforcement procedures now include warnings and throttling in addition to suspension or termination. You can apply for the Cyber Verification Program or Life Sciences Verification Program if your work involves authorized dual-use or high-risk use cases covered by real-time safeguards.
View change record →⚠ The terms will apply as written, and physical action deployments that do not implement independent hardware safety controls and qualified human observation are treated as AUP violations subject to access suspension or termination.
Cross-platform context
See how other platforms handle Physical Action Independent Safety Controls Requirement and similar clauses.
Compare across platforms →"For High-risk Physical Actions, there must be a qualified individual who is able to observe the equipment's operation and stop it at any time, and the equipment must stop or hold a safe state when that individual intervenes or when connection to our services is lost. Any operating limits to keep the equipment within safe bounds, such as speed, force, reach, temperature, pressure, voltage, energy output, dose, or operating area, must be enforced by the equipment or a controller independent of model output.Excerpt from Anthropic's API Usage Policy
1) REGULATORY LANDSCAPE: This provision engages FDA regulations for software as a medical device (SaMD), FAA regulations for autonomous aircraft systems, OSHA machinery safety standards, NHTSA automated vehicle guidance, and IEC 61508 functional safety standards …
Enforcement risk, jurisdiction flags, contract triggers, and due diligence action items.
Get the weekly coverage letter
Companies change their terms without announcing it. We capture governance documents, compare each new version with the last, and publish what changed, with the source text behind it. One email a week on what we found and what we could not reach.
This provision establishes hardware-level architectural requirements for AI-physical system integrations, mandating that safety limits be enforced outside the model control loop and that fail-safe behavior occur upon loss of connection to Anthropic's services, which has direct implications for robotics, medical device, industrial automation, and autonomous vehicle deployments.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by Anthropic.