Central Development
On June 6, TechCrunch reported that Sriram Krishnan will leave his role as a White House AI adviser and plans to start a new institution focused on shaping AI policy, signaling a shift in how industry expertise interfaces with federal decision-making (TechCrunch). The same day, OpenAI introduced “Lockdown Mode,” a configuration designed to reduce prompt-injection risks when models handle sensitive data—while cautioning that the feature cannot eliminate such vulnerabilities (TechCrunch). Separately, Sam Altman has floated ideas for public ownership in AI, a concept also discussed across the U.S. political spectrum by Donald Trump and Bernie Sanders, according to AP News.
Why It Matters
The personnel move suggests policymaking on frontier models may increasingly be channeled through new institutions rather than embedded White House roles. OpenAI’s security configuration underscores that model-integrity risks remain technically persistent even as vendors harden deployments. The reemergence of public-ownership ideas broadens the policy menu under active discussion, from safety standards to potential structural or governance interventions.
Perspective
These developments point to a fragmented but intensifying governance landscape: private actors are rolling out mitigations even as policy entrepreneurs set up new venues of influence. Security concerns are not limited to software: WIRED highlighted that crypto-funded peptide labs in China raise transparency and biosafety questions (WIRED). The outlet also reported claims that Anthropic technology has been involved in NSA-affiliated hacking efforts, illustrating the blurred lines between commercial AI and national-security operations (WIRED).
What to Watch
Details on Krishnan’s new institution: backers, governance model, and whether it partners with federal agencies.
- Whether federal guidance or procurement standards reference tools akin to OpenAI’s Lockdown Mode for sensitive deployments.
- If public-ownership concepts move into hearings, draft bills, or agency studies—and how they’re scoped (model access, data, or infrastructure).
- Follow-on reporting validating or disputing WIRED’s security claims and any policy responses tied to export controls or cybersecurity directives.
