OpenAI's Agents SDK v0.22.1 adds built-in guardrails that block MCP tool calls before they execute, plus configurable sandbox isolation, so agent builders can enforce a safety policy without writing custom code around every tool call. Try: OpenAI's Agents SDK v0.22.1 adds built-in guardrails that block MCP tool calls before they execute, plus configurable sandbox isolation, so agent builders can enforce a safety policy without writing custom code around every tool call. Read: SemiAnalysis's first third-party benchmarks put Google's TPUv7 Ironwood up to 50% ahead of Nvidia's B200/B300 on inference performance per dollar, based on a direct comparison of internal and external TCO. Read: Uber's DragonCrawl has LLMs read mobile app screens and decide what to tap the way a human tester would, adapting to UI changes across 50+ languages without script maintenance. Listen: Lenny's Newsletter interviews the xAI engineer who took Grok Bot from first commit to public launch in about seven weeks, including why the team built it standalone instead of inside Cursor. Read: Google DeepMind's AlphaGenome Atlas precomputes the regulatory impact of all 9 billion possible single-nucleotide variants in the human genome as a queryable, 1-petabyte dataset. Read: OpenAI chief scientist Jakub Pachocki argues the strongest case for continued fast capability scaling is building defensive AI to secure infrastructure and counter rogue agents in real time.
OpenAI Released a Model So Powerful Their Own Safety Team Called It a Critical Risk