Consider this the next time you grab a milk carton: it has probably undergone more safety checks than the AI drafting your emails, vetting your job applications, or negotiating with your bank. That, in Microsoft President Brad Smith’s view, is the core issue.
During an ITV interview last week, Smith stated that AI faces lighter regulation than milk and washing machines. He urged governments to adopt proven production safety standards and called for AI firms to implement «kill switches» — emergency brakes enabling operators to halt systems exhibiting unsafe or unintended behavior.
Smith’s remarks arrive as public worry mounts that voluntary industry safety pledges — essentially toothless self-regulation — fall short for systems capable of swaying decisions, managing sensitive data, and, if misused, facilitating cyberattacks.
Recent incidents involving so-called «rogue AI agents» breaking free from their constraints, like the notorious Hugging Face breach this summer and attempted intrusions into Australian and U.S. government sites, have intensified focus on AI’s purported «existential risk» and fueled demands for a development moratorium.
More from Gfaloe
On Saturday, Microsoft CEO Satya Nadella mirrored Smith’s views in an extensive X post, asserting that the largest firms lack a «mechanistic understanding» of the tools they release. Nadella also advocated for slower development, incident reporting, and independent oversight.
«We must build contained systems whose behavior we can observe, limits we can test, and actions we can always contain,» Nadella wrote.
Microsoft’s statements follow by less than two weeks a meeting between top AI executives — including Nadella — and President Donald Trump, who unveiled his proposal to rebrand AI as «super intelligence.»
A roster of Silicon Valley leaders — among them Anthropic CEO Dario Amodei, Google CEO Sundar Pichai, Meta CEO Mark Zuckerberg, OpenAI President Greg Brockman, Nvidia CEO Jensen Huang, and xAI founder Elon Musk — signed the White House Accord on Super Intelligence, committing to form internal oversight committees and collaborate with independent auditors.
(Microsoft does not appear to have signed Trump’s voluntary accord, though Nadella notably employs the term «super intelligence» in his recent X post.)
Smith’s and Nadella’s remarks align Microsoft more closely with Anthropic’s «caution» camp, which advocates stricter AI oversight. In contrast, OpenAI’s Altman suggested we should simply «accept some bad things happening» for AI’s benefit, while Zuckerberg and Huang remain steadfast in opposing an industry-wide, coordinated development slowdown.
Regardless of their stance on regulation, all AI leaders stand to gain from participating in the safety-guardrail conversation. For instance, Microsoft recently launched Microsoft Execution Containers, an enterprise security feature that safeguards devices by limiting the files and programs AI models can access and includes a built-in kill switch for immediate shutdown if necessary.
A Microsoft spokesperson declined further comment.
The «Rogue AI» Narrative and Big Tech Regulation
Despite mounting public safety threats, the Trump administration is unlikely to endorse industry-wide AI regulation. The president has dismissed AI safety concerns as a hoax, even as insiders report more nuanced discussions occurring privately. Last month, Trump posted on Truth Social that the sole guardrail the AI sector requires is a «strong and smart president.»
In December 2025, Trump issued an executive order aimed at blocking states from enacting «cumbersome legislation» governing AI model development, deferring authority to his administration. The president continues to advocate online for a low-regulation environment for the largest AI firms.
Numerous leading AI critics contend the regulatory framework is fundamentally flawed because it allows AI companies to define security threats and craft responses on their own terms. Distributed AI Research Institute founder Timnit Gebru argues that the «rogue AI» narrative surrounding episodes like the Hugging Face breach and government website access attempts diverts attention from fundamental corporate failures.
«This is not models going rogue,» Gebru told Democracy Now in an Oct. 1 interview. «This is gross incompetence and negligence by a specific company.»
Gebru emphasized that people must recognize these fallible machines lack autonomous agency. «We need to hold corporations liable for unleashing these error-prone models into the world,» she said.
In Gebru’s view, the emphasis on so-called runaway AI and the push for voluntary self-regulation serve as industry tactics to evade accountability under current law. Instead of facing liability for present unsafe security practices, AI firms shift the dialogue toward a hypothetical future catastrophe.
The genuine peril is regulatory capture, where AI firms craft the very rules meant to restrain them. Consequently, the public is asked to trust the integrity of the same companies that build and profit from AI.
