Panel Examines AI Distillation Claims, Ad-Data Surveillance and Cyber Strike AI Firewall Access

Summary

The panel reviews Anthropic's report of large-scale model distillation by three Chinese AI labs, law enforcement purchases of real-time-bidding ad identifiers, and reports that Cyber Strike AI enabled access to 648 firewalls across 55 countries.

Key points

• Anthropic reported that Deep Seek, Moonshot, and Miniax used more than 16 million queries and thousands of fraudulent accounts to obtain model outputs for training their own models.
• Panelists debated whether distillation using a paid AI service constitutes theft or misuse under its terms, and discussed prompt patterns used to get past guardrails.
• A report discussed by the panel said US Customs and Border Protection, ICE, and other law enforcement agencies purchase real-time-bidding advertising IDs to track people and locations.
• The panel discussed privacy and surveillance concerns arising from the commercial collection and use of location and advertising data.
• Cyber Strike AI was described as an open-source offensive AI security platform that can speed up reconnaissance and initial access; its developer was described as having documented ties to China.
• The panel said use of the tool was associated with access to 648 firewalls across 55 countries, where exposed systems, weak authentication, and missing MFA were among the security issues discussed.
• Panelists emphasized basic defenses including limiting internet exposure and strengthening authentication and access controls.

People

Vendors

Tools

Countries

Industries