Anthropic has announced comprehensive measures to prevent misuse of its AI tools during the 2024 U.S. election cycle, the first to see widespread availability of generative AI.
In a blog post dated October 8, Anthropic detailed its strategy to combat potential election-related abuses of its AI model, Claude. The company has implemented strict policies, enhanced detection systems, and redirects to authoritative voting information.
"Since July 2023, we have taken concrete steps to help detect and mitigate against the potential misuse of our tools and to direct users to authoritative election information," the company stated.
Anthropic updated its Usage Policy in May to prohibit the use of its products for political campaigning and lobbying. The policy also bans using Claude to generate misinformation on election laws, candidates, and other related topics, or to target voting machines or obstruct the counting or certification of votes.
To enforce these policies, Anthropic has deployed automated systems and human review processes. The company is also collaborating with Amazon Web Services (AWS) and Google Cloud Platform (GCP) to detect and mitigate election-related harms from users accessing Anthropic models on those platforms.
Vulnerability testing is a key component of Anthropic's strategy. The company conducts targeted red-teaming and Policy Vulnerability Testing (PVT) in collaboration with external subject matter experts to identify potential risks. These efforts help Anthropic refine its systems and improve its ability to prevent misuse.
For users seeking election-related information, Anthropic has implemented a redirection feature. When election queries are detected, users are given the option to access TurboVote, a nonpartisan resource from Democracy Works providing up-to-date voting information.
Transparency is another focus of Anthropic's approach. Claude now includes a clear reference to its knowledge cutoff date, helping users understand the limitations of its election-related information.
Anthropic's blog post concludes: "We cannot anticipate every way people might use our models related to elections, but we have and will continue to learn from and iterate on our processes, testing and improving our systems along the way."
The company has engaged with global policymakers, civil society organisations, and others in industry to discuss its election work. Anthropic has also launched an initiative to fund third-party evaluations that effectively measure AI capabilities and risks.