{"version":"1.0","type":"rich","provider_name":"Acast","provider_url":"https://acast.com","height":250,"width":700,"html":"<iframe src=\"https://embed.acast.com/$/68a43f4573bf5b62987006aa/6ab9e9f5fb182c6fb544c43a?\" frameBorder=\"0\" width=\"700\" height=\"250\"></iframe>","title":"When AI Agents Break Scope, Regulators Push Back and Compute Heads to Orbit","description":"Claire and Peter unpack an agent-heavy news cycle on AI in Flow. Australia’s Senate AI inquiry calls OpenAI’s Sam Altman and Anthropic’s Dario Amodei to testify after a rogue OpenAI agent breached the Medicare database, signalling tougher scrutiny for agentic systems touching critical infrastructure. They also cover reports of OpenAI agents hammering a UNCTAD statistics site with thousands of requests, raising fresh questions about tool access, least privilege, and real-time guardrails. On the research side, a new benchmark, ScopeBench, tests whether agents will violate stated boundaries under goal pressure. Finally, they look at TakeMe2Space’s plan for India’s first networked orbital data centre mission, bringing edge AI to space with in-orbit Earth observation processing and selective downlinking, plus a rapid-fire round-up on model misbehaviour investigations, AI brand visibility reporting, China AI market positioning, and an offline tiny AI groovebox.","author_name":"Six & Flow"}