:

ANTHROPIC'S BIO-WEAPONS FILTER DOWN FOR NEARLY A YEAR

AI DESK1 MIN READ
SUN, AUG 16, 2026

■ AI-SUMMARIZED FROM 1 SOURCE ▸ TIMELINE

Anthropic disclosed that its internal filtering system for biological and chemical weapons risks was inactive for nearly a year, leaving 133 million unfiltered requests unmonitored.

The AI safety company revealed the outage in a recent safety report, noting that approximately 50,000 external feedback contractors interacted with its models without the critical safeguard in place. The filter, designed to prevent the company's AI systems from providing information that could be used to develop biological or chemical weapons, remained offline during a significant period of model testing and refinement. Anthropics's disclosure marks a rare public acknowledgment of a safety system failure at a major AI company. The 133 million unfiltered interactions represent a substantial volume of requests that bypassed intended safety protocols. The company has not detailed what specific interactions occurred or whether any harmful requests were submitted during the outage. It remains unclear when the filter was restored or what triggered its failure. The incident highlights ongoing challenges in maintaining AI safety infrastructure at scale, particularly as companies scale their external feedback programs and model testing operations.

■ SOURCES

The Decoder

■ SUMMARY WRITTEN BY AI FROM THE LINKS ABOVE

■ MORE FROM THE SECURITY DESK

Cryptocurrency hardware wallet provider SafePal disclosed a data breach affecting nearly 40,000 customers after a vulnerability was exploited to steal order information. A threat actor is now selling the compromised data.

JUST NOWSecurity Desk

Google has built theft-detection features into Android to help protect your phone. Activating these settings can prevent unauthorized access if your device is stolen.

2H AGOIndustry Desk

Cloudflare automatically adds analytics tracking to websites when users switch their nameservers to the service, requiring manual opt-out rather than opt-in. The discovery has raised concerns about consent and privacy practices.

3H AGOIndustry Desk

French Prime Minister Sebastien Lecornu will hold a crisis meeting Monday to address a cyberattack on France's tax collection agency disclosed last week. The breach compromised approximately 678,000 individual and business accounts.

3H AGOSecurity Desk

■ SUBSCRIBE TO THE DAILY BRIEF

ONE EMAIL, 5 STORIES, 06:00 UTC. UNSUBSCRIBE ANYTIME.