Anthropic shows AI systems can self-improve on misalignment benchmarks

OpenAI has launched the Rosalind Academic Plan, providing sponsored access to GPT-Rosalind and associated tools for eligible academic research labs focused on life sciences. The program aims to democratize access to advanced AI capabilities for researchers who might otherwise lack resources. The initiative targets institutions conducting work in biology, chemistry, and related fields where computational tools can accelerate discovery.

Anthropic has granted METR researchers employee-level access to its systems, allowing independent assessment of the AI model maker's risks, incidents, and practices. METR will publish its own findings without Anthropic's right to block publication. This arrangement represents a shift toward third-party oversight of frontier AI development.

Science Corporation, led by former Neuralink president Max Hodak, has implanted its vision-restoring chip in its first patient with Stargardt disease. The company is developing a brain-computer interface approach to address vision loss.

AI developers are shifting from relying on readable 'chains of thought' to understand model decision-making toward mechanistic interpretability, a field that examines the internal workings of AI systems. Leading AI companies have established mechanistic interpretability teams, and startups like Goodfire AI have emerged to help decode AI decisions. Goodfire's chief scientist Tom McGrath discussed the approach in a recent interview, highlighting growing recognition that understanding how AI models actually function internally is critical as their reasoning becomes harder to parse.