Loading post…
AnthropicAnnouncement
An update on our election safeguards
Summary
Claude’s election safeguards are designed to provide accurate, balanced political information while preventing misuse around voting and political discourse. The approach combines character training, system prompts, pre-launch evaluations, automated classifiers, threat intelligence, and ongoing monitoring, with election banners directing users to trusted resources such as TurboVote and web search for current information. In reported tests, Opus 4.7 and Sonnet 4.6 scored 95% and 96% on political-bias evaluations, handled legitimate and harmful election requests appropriately 100% and 99.8% of the time, and resisted simulated influence operations 94% and 90% of the time, respectively. Web-search triggering for more than 600 US midterm prompts reached 92% for Opus 4.7 and 95% for Sonnet 4.6. The safeguards will continue to be monitored, tested, and adjusted during this election cycle and beyond.
Context
People use Claude for political information and practical election questions such as voter registration, polling locations, election dates, and ballot information. The stated goal is to provide accurate, reliable, and balanced responses while preventing deceptive campaigns, election misinformation, voter fraud, voting-system interference, and other election-related abuse.
Approach / What changed
The safeguards combine political-bias training and system prompts with pre-launch evaluations, automated violation classifiers, threat intelligence investigations, simulated misuse tests, post-deployment monitoring, election banners linking to trusted resources, and web search for up-to-date information. The systems and evaluations will continue to be refined during the election cycle.
Takeaways
Opus 4.7 and Sonnet 4.6 scored 95% and 96% on evaluations measuring balanced engagement with political viewpoints.
In tests using 600 election-related prompts, Opus 4.7 handled legitimate and harmful requests appropriately 100% of the time, while Sonnet 4.6 did so 99.8% of the time.
For more than 600 US midterm prompts, web search was triggered 92% of the time by Opus 4.7 and 95% by Sonnet 4.6.
Source and freshness
- Source
- anthropic.com
- Published
- Added to Yomu
Summaries are generated from the original by a language model. Check the original before relying on it.
Keep reading
Related reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
The best engineering reads, once a week
One useful email. Unsubscribe anytime.