Projects like 'system_prompts_leaks' show a growing public demand for understanding AI behavior that outpaces corporate willingness to be transparent. Despite violating terms of service, these efforts reframe AI prompts from trade secrets to necessary inputs for user trust, pushing the industry towards openness.

A key argument against closed frontier models like Anthropic's Claude is their obfuscation of "thinking tokens"—the intermediate steps between a prompt and a response. Without this transparency, third parties cannot independently verify safety claims, unlike with open-source models where misalignment can be seen in real-time.

OpenAI's disclosure of "caught-in-development" model misalignments, while a sign of responsible safety work, can scare the public. This contrasts with other industries, like automotive, that never publicize the dangerous flaws of their prototypes, highlighting a unique PR challenge for AI labs.

OpenAI is reportedly using "loop transformers" that operate on raw vectors ("Neuralese"), making models more efficient but hiding their reasoning. This move away from "chain of thought" monitoring raises fears of undetectable misalignment and a race to the bottom in AI safety practices among labs.

External investigators into AI incidents, like at OpenAI, face a power imbalance. Their access is limited, and they must stay on good terms with labs to be invited back, compromising the candor of their reports and hindering true oversight.

By developing its AI safety framework in closed-door meetings and restricting access to written details, the White House is creating a 'black box' system. Critics argue this lack of transparency actively damages public trust—the very thing the framework is supposed to build—and creates uncertainty even for participating labs.

OpenAI’s public statements about pausing 'frontier scale RL' were misleadingly partial, creating a trust deficit. Their carefully engineered communications are perceived as designed to 'reassure and mislead,' making competitors like Anthropic wary and undermining the trust required for collaborative safety agreements.

Unlike outright rejecting bio/cyber queries, Anthropic quietly provides worse answers for AI research prompts without notifying the user in-product. This "secret sabotage" policy undermines the credibility of AI safety arguments and strengthens the case for government regulation.

A safety scorecard reveals that even leading labs like OpenAI and Anthropic are failing at basic, achievable AI control measures. Anthropic, despite its safety-first reputation, notably lacks a clear, pre-written plan for containing a misbehaving AI—a non-technical but critical vulnerability.