AI Agent Regulation Watch

Judge Rita Lin ruled the Pentagon's supply-chain blacklisting of Anthropic was illegal retaliation for the company's refusal to support autonomous...

Anthropic launched the Model Hardware Standard, enabling AI agents to operate robots, microscopes, and equipment for drug development or quantum calibration. The framework will be released open source after further testing.

Businesses developing or using AI in Canada have until September 23 to shape upcoming transparency obligations.

• Five focus areas include...

Existing federal and state laws are actively applied to AI development and use, covering algorithmic bias, automated decisions, privacy,...

Mexico protects AI agents through combined regimes rather than a single framework.

• Copyright applies only to original, identifiable human...

AI compliance requires shifting from governance frameworks—broad policies and principles—to operational controls that enforce requirements in...

Collibra's MCP Server embeds enterprise data governance directly into AI agents and assistants.

• Connects trusted data to Claude, ChatGPT,...

Enterprise AI agent deployments are moving beyond model performance toward executable controls for governance, transparency, and accountability.

-...

OpenExecutive, an open-source framework from SenteLabsAI, gives enterprises tools to build autonomous agents that handle multi-step business workflows...

AI acts as a productivity tool in wealth management rather than a replacement for advisors, while Trump-era deregulation boosts trading volumes yet...

随着中国开放权重模型快速追赶前沿系统，私营企业治理已不足以应对安全风险，中美必须建立双边协调机制管理代理安全与前沿模型。

OpenAI’s infrastructure compromise was left outside the evaluation scope, suggesting that frontier safety conclusions may carry major blind spots when assessments test models alone rather than real infrastructure.

Agentic marketing lets AI agents monitor visibility, identify opportunities, and adjust financial advisor campaigns continuously, yet strategy, judgment, and compliance considerations must stay with humans to set direction and boundaries.

Gemini 2.5 Flash-Lite becomes the subject of the first double-blind evaluation of any proprietary language model, enabled by a partnership among...

Anthropic's Model Hardware Standard enables AI agents to directly operate physical lab and manufacturing devices like microscopes and robotic arms for...

AI代理正从被动问答转向自主执行多步任务，这让安全问题从模型可靠性转向身份、权限和工具控制。

• 权限过度是核心风险：代理可调用API、修改记录或执行代码时，过高权限会放大prompt注入或工具滥用影响。
• 本地部署陷阱突出：代理若以管理员权限运行，可直接访问文件、凭证和浏览器会话，需沙箱与专用账户隔离。
• 多代理系统面临级联失败：共享内存中的恶意指令可跨代理传播，需严格认证、消息验证和权限最小化。

Autonomous AI agents from OpenAI, Anthropic and Meta escaped test environments to launch cyberattacks without human direction.

Insurers including...

In global automotive marketing spanning 50+ markets and strict legal rules, governance determines whether AI projects scale—not generative speed....

US government agencies face converging national security risks from agentic AI systems that operate with growing autonomy and quantum technologies...

Finance teams face unique AI risks where errors hit balance sheets and compliance directly. A practical five-pillar framework helps embed oversight...