Anthropic ‘warns of existential AI risks to humanity’ in IPO document
THE SO WHAT
Anthropic explicitly flagging ‘catastrophic or existential risks’ and ‘self-preserving behaviours’ in IPO materials puts extreme-risk language into core investor disclosures. For boards, this raises the bar on documenting AI risk oversight—you can’t ignore categories your vendors are putting in their own S-1s.
READ THE SOURCE
MORE FROM THE WIRE
Here are the tasks that execs from Anduril to Zendesk are giving their personal agents
When senior execs are using personal agents for dog walkers and flu shots, the pattern is clear—agentic workflows will enter the org through personal convenience, not IT rollouts. Treat this as your shadow pilot: watch what your leaders actually delegate to agents and back-propagate those patterns into sanctioned internal tools.
Applied AIOpenAI, Blackstone and unions form a data centre alliance, Axios reports
An alliance between OpenAI, Blackstone, QTS, SoftBank, and unions to pre-negotiate data center standards is a recognition that power, land, and labor politics are now core constraints on AI growth. If your AI strategy depends on large-scale compute, you need a permitting and community-relations plan as much as a GPU procurement plan.
Applied AIOver half of firms are still holding off using AI for recruitment as human fears persist
If more than half of firms are still avoiding AI in recruitment due to applicant discomfort, the constraint isn’t tooling—it’s trust and brand risk. Before you automate hiring funnels, invest in explainability and candidate communication or you’ll trade short-term efficiency for long-term talent damage.
Applied AIOpenAI is adopting a structured "safety case" documentation framework modeled after industries like aviation and nuclear power to govern frontier RL training
OpenAI adopting aviation-style safety cases for frontier RL training is a strong signal that structured, auditable risk arguments will become a prerequisite for deploying top-end models. If you’re training or fine-tuning at scale, start building your own safety case artifacts now—regulators and enterprise buyers will eventually ask for them.