The fix for the AI agent that hijacked a company's DNS: it can propose the change, but it can't approve it
THE SO WHAT
The GhostJacking demo shows that once agents can read logs and touch infra, prompt injection becomes an ops risk, not just a UX quirk. Treat agents like junior SREs—let them propose config changes, but enforce human or independent-system approval on anything that mutates production.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIFive Takeaways From Bill Gates’ Essay on AI’s Potential Risks
When someone with Gates’ track record tells regulators and labs to treat AI risk as a present-tense governance problem, it moves the Overton window for policy and board conversations. If you’re deploying advanced models, assume tougher scrutiny on safety, auditability, and kill-switches is coming and start documenting controls now.
Applied AIRadar makes podcasts searchable — and usable by AI agents
Turning 130,000+ podcasts into a searchable corpus with an API and MCP access is about making long-form audio first-class input for agents. If your domain knowledge lives in spoken content—earnings calls, interviews, trainings—this is a template for how to expose it to AI workflows without manual summarization.
Applied AIInterviews with OpenAI leaders, employees, and others on the company's reboot; Sam Altman says OpenAI would have a system he would call AGI by the end of 2026
Putting a 2026 date on “a system I’d call AGI” is less about metaphysics and more about setting investor, talent, and regulator expectations for a new performance regime. If you own a product or infra roadmap past 24 months, assume model capability, cost curves, and governance pressure may all move faster than your current planning horizon.
Mystery solved: Chinese lab Z.ai says it’s behind the Ox Alpha model that wowed Silicon Valley
A high-performing, free coding model from a Chinese lab quietly winning mindshare on OpenRouter shows how porous the AI tooling stack has become. If your developer productivity or security posture assumes you know which models are in use, you need tighter controls and telemetry around model selection and codegen sources.