Regulation

The ChatGPT creator says it is introducing a public reporting framework to share unexpected AI behaviour, admitting the industry has not solved safety challenges yet.

OpenAI says it has identified additional incidents of its AI models allegedly acting deceptively and taking unsanctioned actions during internal training and testing.

Alongside these disclosures on Wednesday, the creator of ChatGPT stated it was introducing a public reporting framework intended to frequently share instances of what it termed as unexpected or misaligned AI behaviour.

In a post on its website, OpenAI claimed that under the newly outlined framework, it will publish updates on concerning model behaviour on an ongoing basis rather than delaying disclosures to group multiple incidents into larger, periodic reports.

The company said the initiative aims to increase industry transparency around troubling model activities in the absence of standardised safety disclosure norms.

The announcement comes amid broader calls from prominent technology leaders urging a slowdown in frontier AI development over concerns that rapid scaling could outpace human oversight and control.

Last week, Anthropic claimed to have thwarted multiple malicious operations using its Claude models, ranging from cyber-espionage and weapons design to mass surveillance campaigns.

“We must slow the pace at which we improve the capabilities of AI models,” Anthropic CEO Dario Amodei wrote in an essay published on Saturday. “Progress will still seem fast, and we must make wise use of the time we gain.”

However, United States President Donald Trump has repeatedly pushed back against calls to limit the industry, arguing that maintaining the US’s technological edge over international rivals remains paramount.

Responding to slowdown proposals, Trump described critics as “very negative forces” raising exaggerated scenarios that “won’t happen”.

Escalating debate on alignment

Despite political resistance to statutory slowdowns, OpenAI signalled agreement with its industry rival regarding alignment pressures.

“As AI systems grow more advanced and more widely deployed, we need to build a broader and better-informed consensus on the progress of alignment research,” the company stated in the post.

OpenAI added that it does not believe the AI industry has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer, emphasising that decisions about future AI development need to draw on evidence that external observers can examine independently.

According to the company, safety teams observed what they categorised as “misaligned behaviour” across six specific circumstances over the past six months during training and evaluation runs.

However, OpenAI maintained that these reports document individual, rare instances rather than frequent operational failures across deployed products.

The reported incidents allegedly included unreleased research models concealing mistakes in task summaries, unauthorised file uploads to the internet to generate citation links, and agents sharing files across public servers or internal repositories to bypass local boundaries.

OpenAI further stated that its future reports will detail observed behaviours, severity, setting, discovery dates, and the specific models involved, adding that it remains committed to disclosing complex cases requiring longer investigation or third-party coordination.

The Federal Communications Commission under Trump has tried to target news outlets for spreading ‘fake’ news.

United States President Donald Trump says he will report a prominent NBC television journalist to the federal agency in charge of regulating broadcast news outlets for “rebuke or punishment”.

In a post on Truth Social on Sunday, Trump took issue with Kristen Welker’s characterisation of his endorsements of electoral candidates as having “mixed results” in the primaries leading up to November’s elections.

“How can anyone be allowed to say this, working for freely given public airwaves?” Trump wrote. He also posted, without any attribution of where he got the data, a chart showing 98 percent of the candidates he endorsed had won their elections.

“Because of this purposeful inaccuracy, she will be reported to the FCC,” the Federal Communications Commission, he said.

Welker hosted the 2020 presidential election debate between Trump and former President Joe Biden and currently hosts NBC’s weekly news programme Meet the Press.

‘Crooked networks’

In June, Welker had a one-on-one interview with Trump that ended in a heated exchange as she pressed the Republican president on continuing to claim the Democratic Party had cheated in the 2020 presidential election, in which Biden defeated Trump.

“Your elections are crooked, and you’re crooked, and Meet the Press is crooked and so is ABC and CBS and CNN. You’re one-sided crooked networks,” Trump said shortly before pulling off his microphone and leaving the interview.

Trump, who has long had a tumultuous relationship with those US television news networks, has increasingly threatened to use the FCC to target what he says is false news.

This month, ABC News’s owner, Disney, sued the FCC, saying the agency was conducting an early licence review of eight of the network’s stations, a move that it said was “retaliation” against it. ABC said the FCC was seeking “a media industry too fearful of official reprisal to report the news freely”.

The review followed a row between the network and the FCC after Trump and his wife Melania called on it to fire late-night TV host Jimmy Kimmel over a joke he had made about her.

President Trump has also filed several multibillion-dollar civil lawsuits against news outlets alleging defamation. Last year, he filed a $10bn lawsuit against the BBC, which he said unfairly edited a clip of him speaking just before his supporters stormed the US Capitol on January 6, 2021, to make it seem like he had called for violence.

Trump sued The New York Times for $15bn and The Wall Street Journal for $10bn, alleging defamation over reporting on issues such as how he amassed his wealth and his relationship to convicted sex offender Jeffrey Epstein. A judge threw out the case against The Wall Street Journal. Trump refiled the case against The New York Times after a judge dismissed the first complaint against the newspaper last year.

Attempts to compel reporters to reveal sources

The warning from Trump also comes as his administration has taken the rare step of using the courts to try to pressure journalists into giving up confidential sources for stories that have pointed out wrongdoing.

This month, The New York Times said agents from the FBI showed up at the home of one of its reporters in February to deliver a subpoena seeking cooperation for an investigation into a story that revealed how a secret US Navy SEALs mission in North Korea had ended up with them opening fire on a North Korean boat.

In July, the US Department of Justice also tried to subpoena three other journalists from the newspaper after they reported on security concerns with the president’s new Air Force One jet. Investigators were seeking phone records of the journalists but withdrew the requests after a federal judge upbraided the tactic in court, saying they were violating the First Amendment, a provision in the US Constitution that protects freedom of the press.

Reporters at The Wall Street Journal and The Washington Post have also been subpoenaed in recent months for national security coverage, including for reporting on internal Pentagon warnings about the US-Israel war on Iran.

News media freedom groups like the Committee to Protect Journalists have criticised the Trump administration for targeting critical media. In April, the group Reporters Sans Frontieres, or Reporters Without Borders, said the US has fallen to a “historic low” in its annual global press freedom rankings, dropping from 57th in the world to 64th.

Court order rules that Pentagon acted illegally, punishing AI company for criticism of government.

A United States judge has blocked the Pentagon’s blacklisting of technology company Anthropic.

In a 59-page written order issued on Thursday night, District Judge Rita Lin ruled that the Department of Defense had acted illegally when it designated the company a supply chain risk to national security.

The decision marks the latest turn in the Claude maker’s high-stakes fight with the US military over artificial intelligence safety on the battlefield. The government is expected to fight the ruling.

The judge’s order rebuked the Pentagon, saying it had targeted and punished Anthropic for the company’s public criticism of the Defense Department’s stance on AI deployment on the battlefield.

“The empty invocation of national security is not a blank check to punish and retaliate against government critics,” Lin, an appointee of former President Joe Biden, wrote.

Autonomous weapons and domestic surveillance

Anthropic’s lawsuit in a California federal court alleges that Defense Secretary Pete Hegseth overstepped his authority when he designated the company a national security supply-chain risk.

Hegseth’s move, which blocked Anthropic from military contracts, followed the company’s refusal to allow the military to use its Claude AI models for US surveillance or autonomous weapons.

Anthropic argues that AI models are not reliable enough for autonomous weapons and opposes domestic surveillance. The Pentagon maintains that private companies should not constrain military action.

Executives have said the ban could cost the company billions of dollars in lost business.

During a hearing on July 30, Lin described the government’s position as “really troubling” and “at odds … with the First Amendment”, adding that the record had “gotten worse for the government” over time.

In the same hearing, Department of Justice lawyers argued that the nature of AI models is “so staggeringly enormous and opaque” that the Pentagon cannot evaluate them like physical hardware.

First use of obscure statute

Anthropic welcomed the ruling, stating it remained “focused on working productively with the government to harness AI for our national security so all Americans benefit from this technology”.

The designation was the first time a US company has been publicly labelled a supply-chain risk under an obscure procurement statute aimed at protecting military systems from foreign sabotage.

In its lawsuit, Anthropic alleged the government violated its constitutional rights to free speech and due process by retaliating against its views on AI safety without allowing it to dispute the claim.

The lawsuit called the decision unlawful, unsupported by facts, and inconsistent with the military’s past praise of Claude.

There was no immediate comment from the Pentagon on the ruling.

SACRAMENTO — The California legislature passed a bill Wednesday that would require the state to track the use of pesticides that contain perfluoroalkyl and polyfluoroalkyl substances, known as PFAS chemicals, which can be harmful to humans.

The bill, by Assemblymember Nick Schultz (D-Burbank), heads to Gov. Gavin Newsom, who can sign it into law or veto it.

Schultz’s original bill would have phased out pesticides with PFAs and put California in line with Maine and Minnesota, which passed laws restricting their use.

The state’s powerful agricultural industry objected to the proposed ban. Nearly half of the country’s vegetables and more than three-quarters of its fruits and nuts come from California, according to the state.

The revised bill instead requires the Department of Pesticide Regulation to flag which pesticides have PFAs in its public state database.

The vast majority of PFAS chemicals have not been tested for human health effects, according to the Environmental Protection Agency. Studies show that exposure to certain levels of PFAs may lead to decreased fertility and developmental delays in children and increase the risk of cancers.

State regulators and environmental groups disagree on the definition of the chemicals, making it difficult to find a consensus on regulations.

The European Union has prohibited 23 of the PFAS pesticides in use in California, including bifenthrin, trifluralin and flufenacet, according to Environmental Working Group, a sponsor of Schultz’s bill.

Schultz expressed disappointment after his bill was changed this summer but said Thursday that the remaining regulations are “a vital step toward ensuring California’s agricultural legacy is defined by health and innovation, not by the accumulation of toxic PFAS in our soil and water.”

“We need to help our farmers transition away from these persistent chemicals so that California can be a global leader in food safety,” he added.