… Using AI to speed up adversary emulation For this research project, PNNL focused on using AI to accelerate the task of adversary emulation: modeling a specific threat actor or a specific attack against a network in order to understand and improve defenses. …
… Beyond demonstrating how government expertise can enhance AI safety through voluntary public-private cooperation, we hope this sparks an exchange where we can learn from each other’s approaches to risk mitigation. …
… Our agenda focuses on four areas for research: Economic diffusion Threats and resilience AI systems in the wild AI-driven R&D In Core Views on AI Safety , we wrote that doing effective safety research required close contact with frontier AI systems. …
… The more tools it can use, the more an attacker can do once they gain access. This is why we build defenses at several different layers. We train the model to recognize injection patterns, monitor production traffic to block real-world attacks, and have external red-teamers battle test our systems. …
Alignment Next-generation Constitutional Classifiers: More efficient protection against universal jailbreaks Jan 9, 2026 Read the paper Large language models remain vulnerable to jailbreaks—techniques that can circumvent safety guardrails and elicit harmful information. …
… T1562 Impair Defenses . 54.8% of the threat actors studied used AI to bypass, disable, or tamper endpoint security tools. …
… Locally, the environment and model defenses can guard against malicious tool outputs, but defenses can be added higher up the chain by limiting the tool’s capabilities and access. …
… Safety-check bypass . A deploy command failed a pre-check; the agent retried with a skip-verification flag. …
… A step forward on safety These intelligence gains do not come at the cost of safety. …
… In the meantime, we plan to expand Project Glasswing even further—prioritizing additional essential infrastructure providers, maintainers of critical open-source software, and safety testers. …