Copilot users’ photos reach human reviewers uncensored, 404 Media reports
THE SO WHAT
Human contractors seeing uncensored Copilot photos and prompts—many sexual—shows how thin the abstraction is between “AI” and human review. If you’re ingesting user content for model improvement, you need explicit consent flows, content filters, and a clear story on who sees what and why.
READ THE SOURCE
MORE FROM THE WIRE
Applied AISources: OpenAI's ARR is nearing $70B, growing 70%+ since the beginning of Q3, with B2B revenue up 100%+; it added more consumer revenue in Q3 than all of 2025
If OpenAI is really approaching $70B ARR with 70%+ growth since early Q3 and B2B more than doubling, AI infra and assistant spend is no longer experimental line-item territory—it’s core IT and software budget. Expect procurement, security, and CFOs to start treating model vendors like cloud hyperscalers, with multi-year commitments, diversification pressure, and harder ROI scrutiny.
Applied AINvidia launches new AI safety program designed at stopping models escaping their sandboxes
When a major chip vendor starts baking AI safety into both software and hardware, model containment moves from research topic to infra requirement. If you’re building on their stack, expect new guardrail defaults, compliance checklists, and possibly performance tradeoffs you’ll need to account for in deployment plans.
Applied AIPope Leo says AI safety fears are not ‘fake news’, criticises Jensen Huang
When a global religious leader publicly validates AI safety concerns and calls out industry optimism, the debate moves further into mainstream moral and political territory. Expect more non-technical stakeholders—faith groups, NGOs, policymakers—to demand a voice in your AI risk posture and deployment choices.
Applied AIMeta’s Muse AI sent a YouTuber’s address to a stranger
An AI agent handing a stranger a creator’s home address is the nightmare scenario for “let it run your account” products—data access and action scopes are now a board-level risk, not a UX detail. If you’re deploying agents on top of user accounts or marketplaces, you need explicit permissioning, red-team abuse flows, and kill switches before scale.