Here’s why OpenAI is absent from Nvidia’s industry-wide effort to end rogue AI agents
THE SO WHAT
OpenAI staying off the public list for Nvidia’s Open Agent Safety Platform while reportedly collaborating privately underscores how fragmented agent safety governance still is. If you’re building on multiple vendors, you can’t assume a unified standard—design your own cross-platform guardrails and incident playbooks.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIOpenAI’s latest features take direct aim at the app store model
If ChatGPT becomes a primary surface for discovering and running software — for both humans and agents — the distribution tax shifts from app stores to model and agent platforms. Founders should start modeling a world where “install” is replaced by “invoke via assistant,” and where prompt-level placement and ranking matter more than icons on a home screen.
Applied AIInterviews with Anthropic co-founder Christopher Olah and 20 religious and philosophical leaders about their summits to investigate consciousness in Claude
Once a major lab convenes religious and philosophical leaders to discuss whether its model might be conscious, AI governance moves from pure safety and economics into moral philosophy. If you’re deploying advanced models, you now have to plan for values debates and ethics reviews that look a lot more like bioethics than traditional IT risk.
Applied AIOpenAI Will Keep Driving Down Prices, Sam Altman Says
If OpenAI keeps cutting prices while pushing more capable models and agents, the margin structure of AI-native SaaS compresses toward zero for anything that looks like raw inference resale. Operators should assume per-token costs will fall and compete on proprietary data, workflow depth, and distribution—not on passing through API calls.
Applied AIOpenAI’s Dots are always-on AI agents for work
Always-on "second self" agents like Dots shift AI from a tool you invoke to a co-worker that continuously acts on your behalf—coordination, permissions, and audit trails become the hard problems. Before piloting, map which workflows you’re willing to let an agent touch unsupervised and how you’ll log and reverse its actions.