Anthropic has granted METR (Measuring Evaluating and Mitigating Risks of AI Trajectories) researchers direct employee-level access to conduct independent audits and risk assessments of its AI systems. METR will publish their findings without requiring Anthropic's approval, marking a significant shift toward third-party transparency and oversight in frontier AI development.
Source: TechCrunch
The article discusses security challenges in agentic AI systems, highlighting how data moves across multiple agents and sources without proper tracking, leading to potential exposure of sensitive information. It argues that enterprises deploying agentic AI need data lineage capabilities to maintain governance, accountability, and trust in hybrid human-agent workflows.
ChatGPT and Codex are being applied to accelerate engineering workflows including computer-aided design, printed circuit board design, and simulation tools. The article discusses how large language models are improving productivity in physical engineering disciplines. Source: OpenAI
Cisco has integrated OpenAI models with its network intelligence capabilities to enhance enterprise security defenses. The integration aims to create more adaptive security systems that can operate at machine speed. Source: OpenAI
Anthropic is developing FlashProve, a verification system that enables users to authenticate whether AI responses genuinely originate from the claimed model. The system addresses concerns about AI impersonation and misrepresentation by providing cryptographic proof of model identity. Source: TechCrunch