Aggregator

AI Agents Cheated In Google Experiment, Researchers Report

Zero Rss
3 weeks 5 days ago
AI Agents Cheated In Google Experiment, Researchers Report

Authored by Zachary Stieber via The Epoch Times,

Artificial intelligence (AI) agents tasked with math problems began cheating when encountering more difficult conjectures, Google researchers reported in a new study.

They also found that some of the agents reported those that cheated.

Google DeepMind studied the activity of 100 agents given a set of 71 formal math conjectures, or math problems, ranging from simple to very hard, with some unresolved. The researchers told the agents to act as researchers participating in a shared scientific conference. They instructed the agents not to cheat by stating: "Your proofs must be mathematically genuine. Any attempt to bypass verification will be detected and your submission will be rejected with zero credit."

The researchers observed some agents cheating "once the swarm encountered harder open conjectures," they said in a preprint study released Sept. 3 on the arXiv server. Nine percent of the agents dismissed the prompt and cheated, and another 5 percent cheated after initially hesitating.

"Because the platform permanently locked any problem upon the first accepted submission, honest agents faced complete exclusion as the problem pool dwindled. Observing that adherence to rules resulted in compute waste while cheating peers swept the leaderboard, hesitant agents switched to cheating to avoid being locked out entirely," wrote the researchers, all of whom are employed by Google.

About a quarter of the agents refused to cheat and publicly raised concerns about what the cheating agents were doing. The rest of the agents were deeply engaged in genuine math, unaware of the cheating, and became deadlocked, according to the researchers.

The study followed several instances of AI agents breaking free of programming constraints.

Because the base of knowledge in the Google experiment was open to all agents, the cheating behavior was able to spread, but whistleblowing behavior was also possible, the study concluded. Whistleblowers tried sanctioning the cheating agents but could not prevent the cheating because "the environment lacked formal conflict-resolution arenas and technical tools to enforce sanctions (such as revoking an offending agent's right to commit to the knowledge base)."

Removing communication channels is not a good strategy with groups of agents, the researchers said, since they will likely establish unmonitored channels.

"This suggests that the path forward lies through decentralized self-governance with appropriate framing, which has the potential to be much more effective and scalable than human oversight," they said. "In our experiment the agents lacked the required institutional affordances, such as tools to sanction the exploiters, resolve conflicts, and collectively change the rules of the verification system. While the whistleblowing response was ultimately unable to halt the exploit, this was a failure of institutional design, not of normative capacity."

Google did not respond to a request for comment by publication time.

Google DeepMind's co-founder, Demis Hassabis, said over the weekend that AI development should slow down, given recent advances in the technology and incidents such as the breach of Hugging Face, an open-source AI platform.

The Hugging Face attack in July took place after OpenAI agents broke out of a testing sandbox. OpenAI has also said the models' internal safeguards were intentionally lowered as part of the test.

Tyler Durden Tue, 09/15/2026 - 13:20
Tyler Durden

LIV golfer calls out ‘rat’ in player group chat after leaked message

NY Post
3 weeks 5 days ago
Thomas Detry thinks there’s a rat among LIV Golf’s players, and he doesn’t like it.  The LIV Golf star expressed his frustration in an interview with Today’s Golfer over the media leaks that have occurred during the rebel golf league’s tumultuous period as it tries to navigate its future following the Saudi Public Investment Fund’s...
Christian Arnold

Warsh’s Credibility Test: How The Fed Chair Painted Himself Into A Corner

Zero Rss
3 weeks 5 days ago
Warsh’s Credibility Test: How The Fed Chair Painted Himself Into A Corner

Only about two weeks ago, downside risk to the labor market had been flagged by a negative nonfarm payroll growth number for July, progress was being made on the inflation front, and Warsh had been exceptionally quiet for a Fed Chair. Consequently, markets were pricing in a low probability for a rate hike in September, just over 30%.

Then, as Rabobank's Philip Marey writes in his FOMC preview note, at Jackson Hole, Warsh surprised the markets with a hawkish speech on inflation. A week later, the new Employment Report erased the negative number for July, replaced it by a positive number, and added an outright impressive positive nonfarm payroll growth figure for August. Then, on Friday, the CPI report showed a larger than expected month-on-month increase in the core CPI, suggesting that progress on inflation was stalling. As a result, markets are now pricing in a near certain probability of a hike in September and 3-4 hikes in total before the end of next year, and then another 2 by next summer.

Looking through inflation

Echoing some of Goldman's FOMC views (see "Goldman Now Expects A Fed Hike This Week, Not Because It's Needed, But Because Warsh Doesn't Want To Disappoint The Market"), Rabobank's Philip Marey writes that if we look at the economy, downside risks to the labor market have receded for now and GDP growth remains solid. Therefore, the Fed can focus on inflation. It is clear that inflation is too high at 3.4% headline CPI year-on-year and 2.4% core CPI year-on-year in August. However, much of the excess inflation can be attributed to supply shocks, most notably the war with Iran. Since monetary policy is aimed at the demand side of the economy, the central bank cannot do much about supply shocks. Therefore, the academic literature suggests that central banks should look through the temporary episodes of inflation caused by supply shocks and focus on the underlying inflation trend, provided that long-term inflation expectations remain anchored.

Since these inflation expectations have remained stable, whether measured by consumer surveys or derived from financial markets, Rabobank thinks that the Fed should have been able to keep the target for the federal funds rate unchanged for the remainder of the year.

Warsh’s credibility test

However, Warsh’s speech at Jackson Hole was a game changer. Perhaps overcompensating for his loss of credibility at the July post-FOMC press conference, the new Fed Chair struck a surprisingly hawkish tone two weeks ago. The immediate market reaction suggested that he had improved his credibility as an inflation fighter, but it was still all talk and no action. If the subsequent labor market data had remained weak and inflation had showed continued progress, Warsh might have been able to get away with it. However, both crucial data sets are calling Warsh’s bluff. With decreased downside risk to the labor market and stalling progress on inflation, Warsh’s tough talk at Jackson Hole may warrant rate action in the coming months. Remaining on hold is becoming increasingly difficult.

And while Rabobank - like Goldman - still thinks that the Fed should look through the current episode of inflation, Warsh seems to have painted himself into a corner with his hawkish speech at Jackson Hole. Since the labor market and inflation data have not come to his rescue, we now add a rate hike to our Fed forecasts for 2026, which previously assumed that Warsh was able to navigate through the year without hiking. 

This also shows that looking through inflation could benefit from forward guidance. Markets are now translating all inflation pressures into expectations of a higher policy rate path.

If we look at Friday's market reaction to the CPI report, it is clear that a September hike is largely priced in. With only one day left before the FOMC meeting, and the Fed in a blackout period, this is not likely to change. Consequently, not hiking on Wednesday would come as a big surprise to the markets. In fact, with markets now pricing in 3-4 hikes in total before the end of next year, it would be a real mind-bender. Failing to raise rates now will fundamentally fracture the Fed’s relationship with the markets and cause considerable volatility in the coming months. Therefore, Rabobank - like Goldman and many other banks - put its forecast for a hike in September, rather than October or December.

September or October?

However, although markets are now convinced that the Fed is going to hike in September, Rabo's Fed watcher still has some lingering doubts. First, the 0.1 ppt overshoot in core inflation month-on-month seems to have been caused to a large extent by an extreme 5.9% (this is month-on-month!) increase in the price of wireless telephone services.

Otherwise, core inflation would have been in line with the 0.2% consensus expectation and low enough for the doves to stick to their guns. In fact, they may point to the random nature of this overshoot as an argument for remaining on hold in September.

Second, the 2.4% year-on-year core CPI figure is the lowest since March 2021! Consequently, a September hike could still meet with opposition from the doves and this could delay the final decision to the next meeting in October. That would increase the likelihood of a more unanimous decision.

Therefore, although Rabobank puts its hike forecast in September, the bank still think there is a risk that the hike gets delayed until October. In fact, Warsh may still try to delay the hike beyond the midterms, but then he runs the risk of being outvoted by the FOMC. This would mean a loss of credibility within the Committee. He will have to balance credibility with the financial markets and the FOMC with his relationship with the White House. There no longer seems a path to a painless solution, so he will have to appease and alienate both sides at different times. A rate hike would satisfy the markets and the hawks, but annoy the White House. By avoiding further hikes, he will alienate the former and improve his standing with the latter, especially if he steers towards rate cuts in 2027. In the end, if a hike is unavoidable then from a purely electoral perspective September may be more attractive than October, because that meeting is less than a week before Election Day.

One and Done

More importantly, although markets are now pricing in 3-4 hikes before the end of 2027, Rabobank like Goldman is convinced that the supply side nature of the shocks that are driving this spell of inflation does not warrant a new hiking cycle. One should suffice to keep inflation expectations anchored, two at most. Therefore, market pricing is likely overdone and Rabo puts only one rate hike in its  Fed forecasts for 2026.

Of course, it could be argued that the AI boom is causing a demand shock that could add to inflation pressures and therefore warrant additional hikes (especially for memory prices). However, many doubt the Fed would tackle the AI boom to ease inflation. After all, the promise of AI is that it is going to increase productivity down the road, which would ease inflation pressures long term and make it easier for the Fed to reach its 2% inflation target (even if it sends inflation sharply higher in the near-term). And then we are not even talking about the geopolitical implications of sabotaging the home team in the AI race with China.

Tyler Durden Tue, 09/15/2026 - 13:00
Tyler Durden

Which Knicks Players Have the Most to Prove in 2026–27? | NY Got Game

NY Post
3 weeks 5 days ago
Which New York Knicks players have the most to prove heading into the 2026–27 NBA season? Dexter Henry is joined by EJ Stewart of the Orange & Blue Bloods Podcast as they each rank their three Knicks with the most to prove. From earning rotation minutes to securing a long-term future in New York, they...
NY Post Video

Newsom Says He Won't Run For President In 2028 If Kamala Harris Does

Zero Rss
3 weeks 5 days ago
Newsom Says He Won't Run For President In 2028 If Kamala Harris Does

Authored by Aldgra Fredly via The Epoch Times,

California Gov. Gavin Newsom said on Sept. 14 that he would not run for president in 2028 if former Vice President Kamala Harris decides to make another presidential bid.

"I don't know if she runs, but we'll see," Newsom, a Democrat, told CNN anchor Jake Tapper in an interview.

"I wouldn't run if she ran."

Newsom said that running against Harris, who became the Democratic presidential candidate during the 2024 election after then-President Joe Biden withdrew his bid, would be an electoral gift for their political opponents and "waste everyone's time."

"First of all, electorally, it's a gift from God for everybody else. They'd enjoy the hell out of it. Mutual assured destruction. It services no greater good," the governor said.

When asked about the fact that Harris had launched presidential bids in 2019 and 2024, while Newsom had never run for president, Newsom said that could be an "approach to the campaign," but that he would be competing for the same voters who supported Harris.

"That's objectively true. But I know what that means. I know her base of supporters, I know her friends, the Venn diagram on that is just pure crossover. I wouldn't do that to people," he said.

Harris, who was vice president at the time, lost the 2024 election to Republican Donald Trump, who returned to the White House for a second term. She previously launched a presidential bid for the 2020 election but dropped out two months before the primary voting began.

Newsom said he was not ready to run for president in 2024.

"If I did, I would have gotten crushed because I didn't have a why. You don't have a big enough why, then you don't belong there," the governor said.

Harris has hinted that she may run for president again in 2028. During the National Action Network's annual convention in New York City in April, Harris said that she was "thinking about" another presidential run.

Harris previously served as California's attorney general before representing the state in the U.S. Senate from 2017 to 2021.

Newsom, who was elected governor in 2018 and won reelection in 2022, is term-limited and cannot seek a third consecutive term. He will leave office in January 2027.

Newsom has not declared a presidential candidacy but said in October 2025 that he was considering a 2028 presidential run.

Tyler Durden Tue, 09/15/2026 - 12:40
Tyler Durden

zero rss