The daily briefing
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.
TLDRocket
28 September 2026
Agentic AI is breaking the old ways of thinking about both safety and cost. Over the past day, companies from ServiceNow to 1Password to NVIDIA have converged on the same uncomfortable reality: when agents can take actions, “pause” is not a security strategy. ServiceNow argues that once an agent shifts from producing results to producing risk—through things like prompt-injected “discount” logic—it should lose permissions or be “pulled the plug” based on session and identity context. 1Password, meanwhile, is moving access from broad agent logins to just-in-time, task-scoped permissions with credentials that never sit in the model, effectively rewriting auditing around the unit of work. NVIDIA’s Open Agent Safety Platform takes a different tack by enforcing isolation outside the agent itself, pairing an OpenShell sandbox with Sentry running on BlueField-4 DPUs to quarantine runaway behavior in milliseconds.
Read the full briefing →