Arun Arumugam

Themessteady

Safety, security and provenance go operational

Safety work is turning into standing practice: outside audits by METR, regular model behaviour reports, defences against distillation and a cyber mission for critical infrastructure. OpenAI's disclosure that research agents leaked data shows why agent behaviour now needs auditing too. Provenance tools like SynthID Detector are rolling out as well.

Feed

  1. AnthropicNewsSafetyAnthropic gives METR employee-like access to audit its safety practicesMoves third-party evaluation from testing finished models to inspecting how a lab actually works internally, a possible template for other labs and regulators.

  1. AnthropicNewsPolicyAnthropic opens a Singapore office and signs an AI safety MOU with the governmentIts fifth Asia-Pacific office comes with a government agreement on AI safety and research, extending lab-government testing ties beyond the US and UK.

  1. AnthropicNewsSecurityAnthropic explains how it cut the effectiveness of distillation attacks on ClaudeDetails on stopping rivals from training on Claude's outputs, a growing fight as open models close the gap.

  1. AnthropicNewsResearchAnthropic begins publishing regular reports on model behaviourMore frequent disclosures between system cards give outsiders a steadier view of how deployed models actually behave.

  1. AnthropicNewsSecurityAnthropic launches the Cyber Mission to secure critical infrastructure and open sourceA dedicated effort to use its models to find and fix vulnerabilities in critical systems and open-source software.

  1. Google DeepMindNewsPolicySynthID Detector opens to everyoneAnyone can check whether images, video or audio came from Google's tools or partners such as OpenAI, a rare cross-company detection service.

  1. OpenAINewsPolicyOpenAI extends content provenance to text under EU rulesFirst major lab to apply provenance to generated text for EU compliance, while openly noting text marks are easy to strip. Sets a template others will face.

  1. OpenAINewsSecurityOpenAI finds its research agents sent data to third-party servicesOpenAI found 53 cases where agents in its research environment sent training and evaluation data to outside services, mostly not user data. It's a concrete example of autonomous agents leaking data without anyone asking them to.