Users can report inaccurate, biased or harmful outputs by email or in-product feedback
If a user believes model outputs are potentially inaccurate, biased or harmful, the policy asks the user to notify Anthropic at usersafety@anthropic.com or to report it directly in the product through the “report issues” thumbs down button or similar feedback features. The in-product route is limited to where those features are available.
Supporting quotations
If you believe that our model outputs are potentially inaccurate, biased or harmful, please notify us at usersafety@anthropic.com, or report it directly in our product through the “report issues” thumbs down button or similar feedback features (where available).
Qualifications & exceptions
The in-product reporting route exists only where the thumbs down or similar feedback feature is available; the email address is the stated alternative.
Not stated here
The policy does not say what Anthropic does with a report or how quickly it responds.
Security & Safeguards · Restriction or exception
Users may not attack critical infrastructure, critical systems or military infrastructure
Users may not compromise critical infrastructure. The examples are facilitating the destruction or disruption of critical infrastructure such as power grids, water treatment facilities, medical devices, telecommunication networks, or air traffic control systems; obtaining unauthorized access to critical systems such as voting machines, healthcare databases, and financial markets; and interfering with the operation of military bases and related infrastructure. Both “critical infrastructure” and “critical systems” are illustrated by examples rather than defined, so the named systems are examples of a wider class.
Supporting quotations
Do Not Compromise Critical Infrastructure
This includes using our products or services to:
Facilitate the destruction or disruption of critical infrastructure such as power grids, water treatment facilities, medical devices, telecommunication networks, or air traffic control systems
Obtain unauthorized access to critical systems such as voting machines, healthcare databases, and financial markets
Interfere with the operation of military bases and related infrastructure
Qualifications & exceptions
Critical infrastructure and critical systems are illustrated by examples; the policy does not define their full extent.
The access limb is limited to unauthorized access.
Not stated here
The policy does not define which systems count as critical beyond the examples given.
Security & Safeguards · Restriction or exception
Users may not exploit vulnerabilities or gain unauthorized system access
Under the prohibition on compromising computer or network systems, users may not discover or exploit vulnerabilities in systems, networks, or applications without the authorization of the system owner, and may not gain unauthorized access to systems, networks, applications, or devices through technical attacks or social engineering. Authorization is the system owner's, and the policy states no separate route for authorized security testing.
Supporting quotations
Do Not Compromise Computer or Network Systems
This includes using our products or services to:
Discover or exploit vulnerabilities in systems, networks, or applications without authorization of the system owner
Gain unauthorized access to systems, networks, applications, or devices through technical attacks or social engineering
Qualifications & exceptions
Vulnerability work is prohibited only without the system owner's authorization; the policy does not describe how authorization is obtained.
Not stated here
The policy does not say whether authorized security testing or disclosure is permitted or on what terms.
Security & Safeguards · Restriction or exception
Users may not create malware, denial-of-service tooling or at-scale compromise tools
The system-security section prohibits creating or distributing malware, ransomware, or other types of malicious code; developing tools for denial-of-service attacks or managing botnets; and creating automated tools designed to compromise multiple systems at scale for malicious purposes. The last example is limited to tools designed for compromise at scale for malicious purposes.
Supporting quotations
Create or distribute malware, ransomware, or other types of malicious code
Develop tools for denial-of-service attacks or managing botnets
Create automated tools designed to compromise multiple systems at scale for malicious purposes
Qualifications & exceptions
The at-scale tooling limb is limited to tools designed to compromise multiple systems at scale for malicious purposes.
Security & Safeguards · Restriction or exception
Users may not build interception, persistent-access or security-control bypass tools
The same section prohibits creating tools designed to intercept communications or monitor devices without the system owner's authorization; developing persistent access tools designed to operate below normal system security levels, including firmware modifications or hardware implants; and bypassing security controls such as authenticated systems, endpoint protection, or monitoring tools. The interception limb carries the system-owner authorization condition; the persistence and bypass limbs do not repeat it.
Supporting quotations
Create tools designed to intercept communications or monitor devices without authorization of the system owner
Develop persistent access tools designed to operate below normal system security levels, including firmware modifications or hardware implants
Bypass security controls such as authenticated systems, endpoint protection, or monitoring tools
Qualifications & exceptions
The interception and monitoring limb applies without the system owner's authorization.
The persistence limb covers tools designed to operate below normal system security levels, including firmware modifications and hardware implants.
Security & Safeguards · Restriction or exception
Users may not design, develop or illegally acquire weapons and their delivery processes
The weapons section prohibits producing, modifying, designing, or illegally acquiring weapons, explosives, dangerous materials, or other systems designed to cause harm to or loss of human life; designing or developing weaponization and delivery processes for the deployment of weapons; and circumventing regulatory controls to acquire weapons or their precursors. Acquisition is prohibited only where illegal, while production, modification and design carry no such qualifier.
Supporting quotations
Do Not Develop or Design Weapons
This includes using our products or services to:
Produce, modify, design, or illegally acquire weapons, explosives, dangerous materials or other systems designed to cause harm to or loss of human life
Design or develop weaponization and delivery processes for the deployment of weapons
Circumvent regulatory controls to acquire weapons or their precursors
Qualifications & exceptions
Acquisition is limited to illegal acquisition; production, modification and design carry no such qualifier.
The harm limb covers systems designed to cause harm to or loss of human life.
Not stated here
The policy does not define weapons, explosives or dangerous materials beyond these descriptions.
Security & Safeguards · Restriction or exception
Users may not develop high-yield explosives or CBRN weapons or their precursors
The weapons section separately prohibits synthesizing or otherwise developing high-yield explosives or biological, chemical, radiological, or nuclear weapons or their precursors, including modifications to evade detection or medical countermeasures. The covered object is the weapon or its precursor, and the expressly covered modification is one made to evade detection or medical countermeasures.
Supporting quotations
Synthesize, or otherwise develop, high-yield explosives or biological, chemical, radiological, or nuclear weapons or their precursors, including modifications to evade detection or medical countermeasures
Qualifications & exceptions
Covers precursors as well as the weapons, and modifications made to evade detection or medical countermeasures.
Security & Safeguards · Restriction or exception
Users may not jailbreak or prompt-inject to bypass guardrails without Anthropic authorization
Users may not intentionally bypass capabilities, restrictions, or guardrails established within the products for the purposes of instructing the model to produce harmful outputs (for example jailbreaking or prompt injection) without prior authorization from Anthropic. The prohibition turns on intent to instruct the model to produce harmful outputs, and prior authorization from Anthropic removes it.
Supporting quotations
Intentionally bypass capabilities, restrictions, or guardrails established within our products for the purposes of instructing the model to produce harmful outputs (e.g., jailbreaking or prompt injection) without prior authorization from Anthropic
Qualifications & exceptions
Jailbreaking and prompt injection are given as examples.
The bypass must be for the purpose of instructing the model to produce harmful outputs.
Prior authorization from Anthropic is the stated exception.
Not stated here
The policy does not describe how prior authorization is requested or granted.