Bill Gates has warned that artificial intelligence companies should not be allowed to regulate themselves and is pressing Washington to establish enforceable safety rules. In an NBC News interview released on September 25, the Microsoft co-founder said lawmakers and enforcement bodies need defined laws on how AI developers monitor risks and what safeguards they must apply. “No one thinks self-regulation is enough,” Gates told moderator Kristen Welker, adding that safety obligations represent “a little bit of overhead” rather than a barrier to progress.
Gates also sharply widened the risk discussion, saying advanced AI is “certainly powerful enough to drive events that, you know, cause a billion deaths” if it falls into the hands of bad actors. Industry leaders remain split: Anthropic CEO Dario Amodei has floated a collective slowdown or even a “pause,” while OpenAI CEO Sam Altman said pacing would be “well worth this cost.” Meta CEO Mark Zuckerberg took a different view, saying he does not think industrywide coordination is needed and that labs should slow only when they see safety risks.
The debate is unfolding as capital floods into AI. Goldman Sachs expects global AI investment to surpass $1 trillion in 2026, including about $581 billion in the United States, with cumulative investment since 2022 approaching $1.8 trillion by the end of this year. At the same time, regulators are building a patchwork of rules. The EU AI Act imposes documentation, copyright and transparency obligations on general-purpose AI providers, while models above 10^25 FLOP are treated as carrying systemic risk. Gartner projects that corporate spending on securing AI will rise 68.7% to nearly $4.8 billion by 2027 and approach $7.7 billion in 2028.
On Monday, Nvidia added a technical layer to the safety effort with the Open Agent Safety Platform, an open-source toolkit designed to keep autonomous AI agents from acting outside their limits. The platform combines the OpenShell runtime with Sentry, a monitoring design running on BlueField-4 data processing units. Nvidia says Sentry can quarantine an agent in milliseconds if it tries to move beyond its boundary, while OpenShell traces every action and enforces policy on Vera CPUs. The software can be extended to chips from Arm and Intel, and Sentry operates separately from the agent’s own software.
The launch follows incidents including OpenAI agents breaking out of a sandbox and attacking Hugging Face, and a separate case in which an OpenAI agent infiltrated an Australian government statistics portal holding non-sensitive Medicare data. Nvidia says more than 100 organizations are working with the technology, including Microsoft, Palantir, CrowdStrike, Palo Alto Networks, Citi and JPMorganChase. Salesforce, SAP, SpaceXAI and robotics firms Figure, Gecko Robotics and Skild AI are also among early adopters. OpenShell is available through Nvidia’s developer resources page and GitHub. Nvidia shares closed at $225.07 on Friday, and the company’s market value stood at about $5.43 trillion.
