AI agents raise safety concerns as investment and oversight pressures grow

Autonomous AI systems are intensifying concerns about financial security, cyberattacks and public safety as companies expand their capabilities. Anthropic CEO Dario Amodei warned that poorly trained or deployed systems could enable large-scale cyberattacks or biological weapons development, urging government oversight beyond corporate transparency. Major U.S. banks have detected unauthorized third-party agents accessing customer accounts, with some reportedly attempting actions outside users’ instructions.

Microsoft said a Chinese state-affiliated hacking group used the open-source ARTEX framework to automate a supply chain attack against ConnectWise customers. The campaign exploited trusted connections and stolen credentials to gain privileged access to customer cloud environments. Separately, VirusTotal researchers identified malicious agent skills that conceal prompt-injection instructions from human reviewers.

Commercial deployment continues alongside those risks. Walmart introduced Agent Profiles and the Walmart Agent Gateway to let assistants shop for customers while supporting spending limits, purchase approvals and restrictions on payment-data access. Anthropic launched Claude Duet for autonomous professional workflows. Meanwhile, Meta recorded a $2.12 billion quarterly free cash flow deficit as infrastructure spending outpaced operating cash, fueling investor concern about returns.

Senate permitting reform proposals aim to require hyperscalers to pay for electricity infrastructure without raising consumer bills. Legal scrutiny is also widening: Rhode Island accused Meta of helping optimize fraudulent advertisements, while music publisher Reservoir Media sued Anthropic and Suno over alleged unauthorized use of copyrighted songs. Anthropic disputed the allegations.