An AI now judges every move Rubrik's agents make, its AI chief said at VB Transform 2026 — but no one's measured if the judge is right
THE SO WHAT
Putting an AI “judge” over AI agents without measuring its accuracy is governance theater, not control. If you’re layering AI on AI, you need evals and metrics for every layer or you’re just adding opaque complexity to your risk surface.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIMulti-turn attacks broke AI models 88% of the time — single-turn testing missed it, Cisco AI security lead warns at VB Transform 2026
If your red-teaming is single-turn, your risk model is fiction — Cisco seeing up to 88.3% failure under 6,986 multi-turn attacks across 15 flagship models means conversational adaptation is the real threat surface. Shift eval budgets from static prompts to adversarial dialogues that mirror how attackers actually work.
Applied AIRunway launches AI model router as generative media gets crowded
Runway’s Media Router is a bet that in generative media, control shifts from owning the model to owning the routing layer — whoever sits in front of a “growing roster” of APIs owns the default. If you’re building media tools, assume your users will expect best-model-per-task routing and design your stack to swap providers, not marry one.
Applied AIThe AI compute gap: Enterprises are buying infrastructure faster than they can measure what it costs
When 107 enterprises are accelerating AI infra spend “well ahead” of their ability to see or steer economics, AI capex is outpacing basic cost accounting. Before you sign the next specialized compute contract, instrument per-workload cost and utilization — or you’ll lock in a multi-year burn rate you can’t defend in Q4 budget reviews.
AI 'kill switch' bill unveiled as OpenAI hack raises alarms
A bipartisan bill to let the US government throttle or shut down “risky” AI models moves existential risk from thought experiment to regulatory lever. If your roadmap depends on frontier models, start mapping how a kill-switch regime would hit your SLAs, failover plans, and vendor diversification.