Meta kills AI photo @-mentions after backlash

TechCrunch

Meta quickly pulled an AI photo-referencing feature from its new Muse Image generator after backlash over non-consensual use concerns, highlighting persistent challenges in deploying AI tools responsibly on social platforms.

Anthropic

Anthropic's new Life Sciences Verification Program enables verified researchers to use Claude models with customized, permissive safeguards for biology and drug discovery work, balancing research enablement with biosecurity through behavioral monitoring and tiered access controls.

Z.ai

Z.ai deployed GLM-5.3-Flash using an AI agent to build production inference infrastructure on 100,000+ Chinese accelerators, achieving 3x performance gains in under two weeks. The work demonstrates early recursive self-improvement capabilities where AI models optimize the systems that run themselves, though human oversight remains critical for objectives and risk assessment.

newsletter.pragmaticengineer.com

OpenAI has become fully dependent on Codex agents for virtually all work, with 90% adoption across non-engineering teams and a 10x increase in code output per engineer. The company has built an automated 'software factory' where agents handle code generation, review, deployment, and production monitoring, fundamentally changing traditional software engineering practices and tooling.

Typesafe AI

TypeSafe AI launched Jev, a specialized AI model optimized for structured decisions and automation workflows rather than text generation, delivering 100-444x improvements in speed and cost over traditional LLMs while eliminating hallucinations through mathematical guarantees.