Researchers say suspected Chinese hackers used open-source AI agents to build an autonomous hacking tool that compromised Taiwanese government websites in July
THE SO WHAT
Autonomous AI agents running recon and break-ins against Taiwanese government sites mark a step-change — offensive cyber is now partially automated and open-source. Security teams need to assume attack volume and sophistication will scale with model access, and start testing defenses against agent-driven campaigns, not just human operators.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIAnthropic’s Watermarking, How It (Probably) Works, Worse Than It Seems
If watermarking is being bolted on primarily to satisfy EU AI law, expect real friction between compliance checkboxes and technical robustness. Operators relying on provenance for safety, IP, or trust should treat current watermarking as one weak signal among many — not a single source of truth.
Applied AIRyanair signs a five-year AI deal with Google Cloud, and a second cloud to fall back on
Ryanair making Gemini Enterprise the operational center of a 35,000-person airline — while explicitly keeping a second cloud in play — is the reference pattern: deep AI integration plus negotiated optionality. If you’re signing multi-year AI platform deals, bake in multi-cloud leverage and exit ramps before agentic workflows become core to how you run the business.
Applied AISpaceXAI launches Grok Bot, an always-on AI agent
An always-on swarm of Grok agents pushes the assistant model toward persistent, ambient automation rather than chat-on-demand. Teams building products should assume users will have background agents monitoring, summarizing, and acting — design your APIs, permissions, and UX for machine clients, not just humans.
Forget chatbot training. AI's next big data grab is about learning how humans work.
The next defensible data moat isn’t text corpora—it’s logs of how real employees click, decide, and resolve edge cases across tools. If you’re not already instrumenting workflows and capturing high-quality interaction traces, you’re handing future agent performance advantages to whoever is.