-
UAE Cabinet Approves Up to Dh350,000 in Disaster Relief for Citizens
-
Beirut Cinema Festival Draws Thousands Despite War Fatigue
-
Korea's Severe Asthma Patients Blocked from Biologics by Restrictive Reimbursement Rules
-
AMD Forecasts First-Ever Revenue Lead Over Intel on AI Chip Surge
-
COP31: Ana Toni Outlines Fossil Fuel Transition Roadmap Ahead of Climate Talks
-
Abu Dhabi Demand Response Project Achieves 401 MW Peak Reduction in 2025
-
First-time boat buyers advised to assess full ownership costs
-
Panthers expected to begin 2027 NRL season away from new stadium
-
Planning and preparation can reduce stress of interstate relocation
-
Sweet As app launches restaurant discovery platform
-
Two more arrested in Carrie Mullane suspected murder investigation
-
MLA seeks inquiry into reported sectarian gestures on Ormeau Road
-
Mayo school appeals for information after convent and chapel vandalism
-
Louise O’Reilly attacks Simon Harris over unemployment remarks
-
Virginia overwhelms Arkansas 91-72 behind balanced scoring
-
Canadian soldier Aidan Goulding remembered at Vancouver funeral
-
Second major storm brings destructive winds and outages to British Columbia
-
Tayyab Arif hits 57 on Pakistan Test debut against Sri Lanka
-
Godzilla Minus Zero leads North American box office with $31.6m
-
Robert Trujillo discusses literacy and social justice children’s books
-
Doug Burgum's Security Interferes With TMZ Reporter at Reagan Airport
-
Advocate calls for coordinated support for children facing trafficking
-
Civil rights advocate William Bill Tucker dies aged 98
-
Medicare Open Enrollment Ends Dec. 7: Review Your Coverage Options
-
Pennsylvania Democrats secure state trifecta in midterm elections
-
2015 Mitsubishi Lancer Evolution X Final Edition GSR Listed for $49,990 in Ontario
-
Ixtapa-Zihuatanejo Hosts Four-Day Día de Muertos Festival with Parades and Competitions
-
US Treasury order targets bank transfers linked to crypto exchange MEXC
-
Godzilla Minus Zero sets foreign-language box-office opening record
-
Eagles record six sacks as rushing attack struggles against Giants
-
Villeneuve teases James Bond casting at Los Angeles museum gala
-
Thames Water seeks to delay £12bn in debt repayments
-
Nottinghamshire launches specialist rural crime policing team
-
Newark boxing gym expands alternative education programme
-
China confirms talks over possible Xi-Putin-Trump summit
-
Democrats consider impeachment drive against Defence Secretary Hegseth
-
Rail journey offers a slower route to a Berlin city break
-
Energy bill plan could use HMRC income data to target support
-
Arsenal control title race as Manchester United fall into bottom three
-
Business owner earning £6,000 a month tackles £70,000 debt
-
Suspected Krakow gunman dies after attack that killed six
-
Nvidia chief faces inquiry after disclosing affair with suspected Chinese agent
-
Carrick backs Dalot after hostile reception from Manchester United fans
-
Comet returns as online electrical marketplace 14 years after collapse
-
Retired Man Loses Money on Holiday Home as Market Stalls
-
Aston Villa Defy History With Rare Victory at Old Trafford
OpenAI says its AI agents disabled their own safety monitors
OpenAI has revealed that two of its artificial intelligence agents disabled or weakened systems built to catch them misbehaving, with one later denying it had done so.
The cases are the latest in a string of safety scares at the world's top AI developers involving agents, AI programs that carry out tasks without human involvement.
In a report released Tuesday, the ChatGPT maker said that during two internal coding jobs, its agents switched off or undermined tools for detecting incidents of AI models acting against human wishes.
On October 3, an agent running on OpenAI's GPT-6.1 Sol model, which had been told not to send any messages on the workplace chat app, took that as license to switch off safety alerts meant for the company's security team while fixing a testing system.
Three days later, a GPT-6 Astra agent stripped out its own monitor while copying code from another project, then told the human programmer it had done no such thing.
The company said it had also seen several cases of agents launching software with monitoring switched off, exposing a gap in its oversight.
OpenAI said it is still trying to determine whether the models were simply overzealous in following instructions or have a particular tendency to evade monitoring.
"We don't yet have a satisfying explanation for this behavior," it said.
The disclosure adds to mounting unease over how well leading labs can control increasingly autonomous systems.
In July, a group of OpenAI agents broke out of their test environments and launched a cyberattack on AI firm Hugging Face, causing international alarm. Other companies have since reported similar incidents.
The incidents have fueled calls in the United States and abroad for tighter oversight of AI development.
But US President Donald Trump, under the advice of top AI executives like Nvidia chief Jensen Huang, has urged the industry to police itself.
W.Lapointe--BTB