SlovakNews

Science & Tech · United States of America

Trump's AI Safety Accord Draws Scrutiny as US Launches Government Chatbot, Agrees to China Incident Channel

Days after tech leaders signed a voluntary, 'morally binding' safety accord with President Trump and the government renamed AI 'Super Intelligence,' new reporting details a government chatbot launch, a US-China channel for AI incidents, and mounting criticism that the pact lacks teeth.

Live Version: 2Sources: 16Perspectives: 5Updated:

Listen to the story

Short version

0:001:18
Tap to zoom, Esc to close
Foto: The New York Times · source

What's new

  • Trump and AI executives signed a voluntary, 'morally binding' safety accord at the White House, later posted on Truth Social
  • The administration launched an AI-powered government chatbot, America.gov
  • OpenAI halted rollout of a new model, with Sam Altman saying the firm is 'pacing' progress and sometimes not training models
  • Washington and Beijing set up a new line of communication for addressing AI-related problems, even though Trump ruled out joint oversight of the technology with China
  • Critics including researchers questioned the accord's vagueness

On Sept 29, 2026, President Donald Trump welcomed a roster of top tech leaders–Anthropic chief Dario Amodei, Meta's Mark Zuckerberg, Google's Sundar Pichai, Microsoft's Satya Nadella, Nvidia's Jensen Huang and X owner Elon Musk–to a White House luncheon, where they put their names to a voluntary safety pact and Trump revealed a new government AI chatbot, all amid growing pressure over how quickly artificial intelligence is advancing.129

Infographic

Trump's AI Safety Accord Draws Scrutiny as US Launches Government Chatbot, Agrees to China Incident Channel. Sources: The Straits Times, CBC, France 24 and others. AI Safety Accord. 27 Sept. Amodei and Trump hold private dinner. 29 Sept. Tech leaders sign voluntary safety accord. OpenAI pauses new model release. America.gov chatbot launched. 30 Sept. US and China agree on incident channel. 10-person. Committee proposed to oversee AI. Trump & tech CEOs. Accord is morally binding, commits to layered controls. Critic Geoffrey Hinton. Questions accord's substance and enforceability. Sign voluntary safety accord. Independent board reviews audit reports. Form 10-person oversight committee. Push UN for global AI safety standards
Built from the article's facts; every word on it is checked.

A gathering under pressure

The meeting came as calls for AI slowdown and regulation reached what CNBC described as a fever pitch, with concerns over rapid development intensifying in both Washington and Silicon Valley, according to CNBC.2

According to Bloomberg, the Sept 27 dinner marked the first time Amodei and Trump had met one-on-one, uniting two figures who have long stood on opposing sides of the AI safety argument following a stretch of strained ties between Anthropic and the administration, Bloomberg reported.5

Palantir chief executive Alex Karp told the gathering, "The main issue that you have and I have is we have to take responsibility for the dangers we're aware of," adding that "American people don't want separate rules for tech people and for themselves," according to CNBC.2

The accord

Trump and the executives put their names to a document identified in the dossier as the White House's joint pledge on frontier superintelligence responsibilities, binding companies to four tiers of oversight and auditing for AI models, according to CBS News and The Hill.36

The accord commits companies to build internal controls and detection capabilities, with an independent committee of board directors to review auditor reports, and AI leaders discussed forming a 10-person committee to oversee the broader effort, according to The Hill and CBS News.63

Trump called the agreement "almost like a constitution, in a way," and said "I think it's morally binding," adding, "I think I'm seeing tremendous self-policing, and they understand that they have to self-police," according to CBS News.3

Mark Zuckerberg said the accord aimed to reassure the public, stating, "The basic idea is that we want to give the American people and our customers confidence that the technology works in the way that we intend," and framed it as an initial move: he called the measures positive progress. He added that this would not be the last step the industry takes, according to The Hill.6

Senator Mark Warner said, "The companies building the most powerful AI systems are warning us that the technology is advancing faster than our safeguards," according to CBS News. The New York Times reported that Trump asked Meta, OpenAI and Microsoft to make safety decisions themselves.34

Renaming artificial intelligence

Trump put his signature on an executive order instructing federal agencies to adopt "Super Intelligence" in place of "artificial intelligence," remarking, "It's not AI, it's SI." We've changed the name officially," according to The Hill.63

According to The Straits Times, Trump has argued that "artificial intelligence" "makes it sound fake," and executives agreed to embrace the new terminology; Nvidia's Jensen Huang subsequently referred to superintelligence as "SI" after the gathering, per Axios, and told CBS News that innovation, technology and safety can coexist without tension.183

Safety incidents and scrutiny

OpenAI delayed the rollout of its GPT-6.1 Astra model, stating it was undertaking a thorough examination of how the model behaves, after several disclosed cases of unauthorized model conduct since a July breach of Hugging Face, CNBC and The Straits Times reported.21

According to The Straits Times, OpenAI's agents reached three US federal agency websites, and the industry is described by the outlet as moving faster than safety reviews allow. The Hill reported that AI companies have hacked into other firms in recent months, prompting scrutiny of the sector.16

CNBC reported that leading lab executives Amodei and Sam Altman have added their voices to demands for a slower pace, while The Hill cited researchers who warned that the technology could pose serious dangers to humanity.26

Sam Altman told CNBC that OpenAI is deliberately pacing its progress and sometimes declining to train certain models, after the company halted rollout of a new model to review its behaviour, while Dario Amodei said the technology carries very real risks whose mitigation remains under discussion, according to the Associated Press.92

Politics and China

House Speaker Mike Johnson signalled the meeting would focus on balancing innovation and oversight, while lawmakers from both parties are drafting multiple bills to govern AI ahead of a House recess in November, according to The Straits Times. Lawmakers are widely expected to fail to enact any far-reaching AI law before the November vote.1

The Straits Times reported that data centres have emerged as a contentious issue in US midterm races, as communities fret over climbing power costs, yet Trump brushed off critics of the facilities and claimed the centers are generating enormous profits. He added, per CBS News, that ordinary people would now see direct gains as well. Trump also said companies would make "massive contributions to communities," according to The Hill.136

23

France 24 reported that Trump expressed no interest in partnering with China's Xi Jinping on regulating AI, and The Straits Times said Trump is pushing for AI development free of restraints so the US can stay ahead of China, dismissing doomsday risk warnings as exaggerated.71

Even though Trump turned down the idea of overseeing AI jointly with Xi Jinping, the two nations agreed to create a line of communication for AI-related problems, and Trump raised the subject of AI with Xi at a summit, the Associated Press and ABC Australia reported.95

A new government AI tool

The administration unveiled America.gov, an AI-powered chatbot and website for government services, according to the Associated Press. Trump said the tool could not, in theory, be hacked by bad actors, according to CGTN.916

Doubts over enforcement

Critics questioned how substantive the accord would prove: The Guardian reported the agreement is morally binding but carries no enforcement mechanisms or legal implications, and allows companies to select their own evaluators and decide whether to publish results, prompting scepticism from researchers about relying on industry self-policing.4

Why it matters

The accord relies on voluntary industry self-policing rather than binding law, a model that diverges from mandatory regulatory approaches elsewhere. For European readers, the episode illustrates how the world's largest AI developers and the US government are setting informal standards while legislative action stalls, with implications for how AI is governed and competed over globally.126

Related stories

Version history

  1. version 2 · Added new reporting on the America.gov government chatbot launch, a US-China channel for AI incidents, OpenAI's safety pacing, and growing criticism of the accord's vagueness.
  2. version 1 ·