Overview
- At Google Cloud Next on Wednesday, Google launched Gemini Enterprise to build and govern autonomous agents, detailed security agents that hunt, detect and fix threats, and confirmed a Wiz partnership for red, blue and green agents that run pen tests, verify breaches and auto‑remediate code or settings.
- Google also introduced TPU 8t for training and TPU 8i for inference with higher compute, lower latency and integrated power management, and it highlighted tools like Agent Studio, cryptographic Agent Identity, agent‑to‑agent orchestration and an inbox for long‑running tasks.
- Anthropic, which says its Mythos model can find and exploit hidden software flaws, is restricting access to vetted partners and coordinating with governments, and it said Tuesday it is investigating reports of unauthorized access to a version of the system.
- Amazon will invest up to $25 billion in Anthropic, reported Tuesday, with an initial $5 billion and more tied to milestones, and Anthropic will use Amazon’s Trainium chips to secure the compute needed for new Claude and Mythos‑era work.
- Meta confirmed an internal program that records mouse moves, clicks, keystrokes and occasional screenshots on work apps to train agents, employees reported no opt‑out on corporate devices, and new research and reports point to AI‑linked layoffs and rising worker strain as adoption surges.