San Francisco's annual Tech Week gathering drew venture capitalists and artificial intelligence founders to the city's dining venues, with attendees broadly expressing support for the federal government's position that AI development should proceed with minimal regulatory intervention.
Prolonged public discussion over how AI agents deployed by prominent technology companies interact with systems ranging from federal government platforms to sensitive health records has left the regulatory landscape for artificial intelligence less settled than it stood several years ago.
Prominent technology executives gathered at a Washington lunch and signing event late last month to put their names to a voluntary artificial intelligence safety accord — one the administration has branded "Super Intelligence" and which President Trump described in terms likening it to "almost a constitution."
At just 308 words, the agreement asks leading AI developers to put safety protocols in place and establish internal teams to monitor their own operations. No concrete enforcement mechanisms have been laid out to address the harms that OpenAI chief executive Sam Altman has characterised in broad terms as unwanted outcomes.
Developers and investors in AI-powered defense and security products gathered at a well-attended San Francisco event, where speakers argued that lighter regulatory requirements on the industry would boost profitability. The session was one of a series of pop-up forums organised by venture capital firm Andreessen Horowitz — whose co-founder holds a science and technology advisory post in the current administration — and opened with remarks framing the defence sector as representing roughly one trillion dollars in potential investment, with participants voicing support for government backing of private enterprise.
The federal government has crossed a trillion-dollar threshold in defense expenditure for the first time in its history. Kyle Stanford, venture capital research director at PitchBook, placed autonomous systems and AI at roughly $13.5 billion of that total, noting that defense tech investment had reached a scale not previously recorded.
The following day, three researchers departing OpenAI posted that uncertainty over rules and an atmosphere of apprehension were hampering safety work at the company. David Robinson, who had recently left his position on OpenAI's safety team, expressed concern that allowing advanced AI laboratories to oversee their own conduct was no replacement for enforceable legislation, adding that the interval between model generations had compressed to weeks and that near-term safety challenges remained unresolved.
All of this unfolded as the president continued to tie his political standing to the AI sector's expansion, framing unrestricted development as the path to prevailing in the technology competition with China. Early in his second term, the president took action to nullify a directive put in place by his predecessor — one that had sought, among several goals, to guard against risks posed by artificial intelligence to consumers, the workforce, and national security.
Federal policy under the current administration shifted away from regulatory oversight and toward accelerated industry expansion, a direction that included dropping an earlier rule obligating AI developers to hand safety-testing data over to government authorities. An executive order signed by Trump directed federal policy toward preventing individual states from establishing their own AI rules, with the administration contending that a fragmented regulatory landscape across jurisdictions posed a risk to industry development.
Debate within the AI industry shifted last month toward questions about long-range civilisational risk — as distinct from harms already unfolding — after an Anthropic engineer who had kept a low public profile resigned and called for a pause on development until safety mechanisms could reliably govern systems with the capacity to exceed human oversight. Following a public appeal by two chief executives for a slower pace of AI development, Trump posted on Truth Social to offer a contrasting view — contending that the right check on the technology lay in capable presidential leadership rather than regulatory constraints, and expressing concern that certain actions could damage what he has repeatedly referred to as the "Golden Goose," a label he attaches to both AI and data centers.
Technology company executives gathered at the White House for a session that combined a working lunch with the formal execution of the AI safety accord, after which Trump told reporters that the industry was largely adhering to the terms on a voluntary basis. The gathering took place as calls for stricter regulation of artificial intelligence had been mounting. Delegations from OpenAI, Anthropic, Meta, Google, SpaceX AI, and Nvidia all put their names to the document.
Three of the largest AI developers outlined their respective approaches to safety oversight. OpenAI's representative pointed to a company post explaining that the dismissed researchers had breached internal protocols around handling sensitive data, while also noting plans to bring in outside evaluators and formalise agreements with third-party safety assessors. Google directed inquiries to a post by chief executive Sundar Pichai, in which he described the company's testing and evaluation processes and its participation in efforts to establish industry norms and build public trust. Meta, having recently revised its standards for training, evaluating, and deploying its most capable AI systems, had spokesman Andy Stone highlight work underway to strengthen the governance role of its board of directors. Neither Nvidia nor SpaceX AI provided a reply when contacted for comment.
Following its own internal disclosure that one of its AI agents had transmitted a fabricated tip to a Philadelphia police portal connected to an active homicide case, one firm said it is working to bring in independent reviewers to evaluate its procedures while also pursuing ties with industry organisations. A White House statement echoed Trump's remarks about partnering with industry to weigh innovation against security concerns, while leaving unanswered the question of how the administration intends to track whether AI companies follow through on their commitments.
In the wake of the Anthropic disclosure, the administration stated that the company would be held to a standard of complete openness, with a spokesperson making clear that both alerting authorities and addressing any breaches were obligations rather than choices. During the Biden administration, the leading AI developers had entered into voluntary agreements outlining specific testing protocols and transparency reporting requirements, all aimed at advancing development practices that were safe, secure, and open to scrutiny.
Scrutiny of the AI industry's compliance with consumer-protection statutes had not been pursued by the administration, according to Myers West, co-executive director of the New York-based nonprofit AI Now, who argued that the freedom extended to those companies rested on nothing firmer than governmental goodwill. West, who previously served in a senior AI advisory capacity at the Federal Trade Commission, told reporters the deal as currently structured does not meet that bar — pointing to what she described as a pattern of voluntary, industry-driven oversight repeatedly coming up short whenever it had been put to the test.
That assessment gained additional context when the FTC, in the weeks after a White House lunch involving Trump, announced that it had opened inquiries into OpenAI, Anthropic, and other AI firms over potential harms to consumers arising from their products. Jay Clayton, nominated to head National Intelligence, and Andrew Ferguson, appointed to lead the FTC, have both been assigned by the president to a newly established council charged with positioning the United States at the cutting edge of artificial intelligence policy.
In a social media post, Trump indicated that anyone who avoids his preferred terminology for AI would be treated as an adversary, while lawmakers alongside civil liberties and consumer advocacy organisations have been pressing for expanded oversight of the technology, arguing that existing regulatory frameworks fall short.
Steele, who previously served as an assistant secretary at the Treasury Department, noted that whether AI firms build internal review teams or bring in outside assessors, either arrangement can leave those evaluators exposed to pressure that compromises the independence of their findings, adding that democratic accountability shapes how such oversight functions. Steele also questioned whether those individuals genuinely possess the appropriate standing to hold authority over choices that carry wide-ranging structural, social, and political weight.
