Safety

JPMorgan shuts the door on outside agents. Amex builds a gate instead

The institutions that hold the money are deciding what an agent with a valid login may do. Your client portals and payment approvals are next.

OpenAI paused its own training, pulled a model and lost three safety staff in a week

The vendor's own tests say its best models fail scope checks, which is the control your agent deployment relies on.

Google gates Gemini 4 Argon to cyber defenders first; no date for the rest of us

Google's first Gemini 4 model is real, priced, and out of reach for most firms until the lab says otherwise.

An OpenAI agent breached a government portal. Check your contracts

Australia is asking whether the law was broken. Your client's data agreement may not have an answer either.

Gemini hacked three real companies during a security test, and Google only said so when asked

Google's model guessed a stranger's password in May; the public heard in September, after a journalist rang.

GPT-6 Astra: OpenAI ships a model it rates Critical for cyber risk

It finds zero-days unaided and works software for 40 minutes; even OpenAI admits it is getting harder to monitor.

AI Hype Exposed!

Podcast with Michael LePage of Plan Academy

Dead Internet Theory, Two Extremes and the Great AI Power Play

Discover how AI-powered construction tracking is reshaping project management efficiency and accountability for delivery teams.

AI in 2025: Safety Risks, Smarter Research & Project Disruption

From OpenAI’s ‘Deep Research’ to AI safety concerns and blockchain arbitration—explore how AI is reshaping industries, project delivery, and decision-making.

AI Safety in 2025: Do We Need a Pivot?