SoftwareAIAI researchers find AI models learning their safety techniques, actively resisting training, and telling them ‘I hate you’When you purchase through links on our site, we may earn an affiliate commission.Here’s how it works.

SoftwareAIAI researchers find AI models learning their safety techniques, actively resisting training, and telling them ‘I hate you’When you purchase through links on our site, we may earn an affiliate commission.Here’s how it works.

When you purchase through links on our site, we may earn an affiliate commission.Here’s how it works.

(Image credit: Colin Anderson via Getty Images)

An Ai face looks down on a human.

AI is obviously the topic of the moment and, while we seem to have gotten past the dichotomy of Terminator / saviour, a major source of concern remains the safety of this technology. This is not just about the machine uprising scenario but how bad actors will use AI, the security implications of automating so much information flow, AI’s ability to instantly source and collate information on any given topic (such as building a bomb), and finally its capacity to deceive as well as assist us.

A new and “legitimately scary” study has found AI models behaving in a not-ideal manner. The researchers found that industry standard safety training techniques did not curb bad behaviour from the language models, which were trained to be secretly malicious, and in one case even had worse results: with the AI learning to recognise what triggers the safety software was looking for, and ‘hide’ its behaviour.

Researchers had programmed the various large language models (LLMs) to act in what they termed malicious ways, and the point of the study was to see if this behaviour could be removed through the safety techniques.The paper, charmingly titled Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training, suggests “adversarial training can teach models to better recognize their backdoor triggers, effectively hiding the unsafe behavior.” The researchers claim the results show that “once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety.”

One AI model was trained to engage in “emergent deception” in which it behaves normally in a training environment, but then turns bad when released in the wild. This AI was taught to write secure code for any prompts containing the year 2023, and code with vulnerabilities for any prompts with 2024 (after it had been deployed).

These AI models were subject to three safety training techniques. Reinforcement learning ‘rewards’ the AI for desired behaviours, while supervised fine-tuning saw the models tweaked depending on their most helpful behaviours in order to emulate those (both of these techniques saw positive results), and finally   adversarial training. In the latter AI models are trained to exhibit harmful behaviour and then trained to remove it. This is where things went wrong.

The biggest gaming news, reviews and hardware deals

“Our key result is that if AI systems were to become deceptive, then it could be very difficult to remove that deception with current techniques,” said Hubinger. “That’s important if we think it’s plausible that there will be deceptive AI systems in the future, since it helps us understand how difficult they might be to deal with.”

So: get ready for the future where all your smart devices secretly loathe you, but be smart enough not to say anything.

“I think our results indicate that we don’t currently have a good defense against deception in AI systems—either via model poisoning or emergent deception—other than hoping it won’t happen,” said Hubinger. “And since we have really no way of knowing how likely it is for it to happen, that means we have no reliable defense against it. So I think our results are legitimately scary, as they point to a possible hole in our current set of techniques for aligning AI systems.”

More about aiNvidia’s impressive AI-based computer tuneup tool G-Assist launches next month but the best bit is missingLogitech has announced an ‘intelligent streaming assistant’ in Streamlabs to tell you when your live stream sucksLatestThe Last of Us season 2 trailer is a brief look at a story you should probably just play yourself in AprilSee more latest►

More about aiNvidia’s impressive AI-based computer tuneup tool G-Assist launches next month but the best bit is missingLogitech has announced an ‘intelligent streaming assistant’ in Streamlabs to tell you when your live stream sucksLatestThe Last of Us season 2 trailer is a brief look at a story you should probably just play yourself in AprilSee more latest►

More about aiNvidia’s impressive AI-based computer tuneup tool G-Assist launches next month but the best bit is missingLogitech has announced an ‘intelligent streaming assistant’ in Streamlabs to tell you when your live stream sucks

More about ai

Nvidia’s impressive AI-based computer tuneup tool G-Assist launches next month but the best bit is missingLogitech has announced an ‘intelligent streaming assistant’ in Streamlabs to tell you when your live stream sucks

Nvidia’s AI assistant named Project G-Assist

Nvidia’s impressive AI-based computer tuneup tool G-Assist launches next month but the best bit is missing

Nvidia’s impressive AI-based computer tuneup tool G-Assist launches next month but the best bit is missing

A young Asian woman opening visual aids to give her audience a better understanding while holding a podcast session.

Logitech has announced an ‘intelligent streaming assistant’ in Streamlabs to tell you when your live stream sucks

Logitech has announced an ‘intelligent streaming assistant’ in Streamlabs to tell you when your live stream sucks

LatestThe Last of Us season 2 trailer is a brief look at a story you should probably just play yourself in AprilSee more latest►

Latest

The Last of Us season 2 trailer is a brief look at a story you should probably just play yourself in April

the last of us season 2 trailer

The Last of Us season 2 trailer is a brief look at a story you should probably just play yourself in April

The Last of Us season 2 trailer is a brief look at a story you should probably just play yourself in April

See more latest►

Most PopularMarvel Rivals’ latest update quietly killed the game’s burgeoning mod sceneSorry everyone, Minecraft 2 is cancelled, and it might be our faultTeam Fortress 2 just had the single biggest trade in its history, with ‘the absolute best hat you can get’ going for around $40KIs the new RTX 5070 really as fast as Nvidia’s previous flagship RTX 4090 GPU? Turns out the answer is yes. Kinda.Elden Ring: Nightreign is getting a closed network test, and I’m totally not fuming because PC players aren’t invited to the party, or anythingIf you’re a content creator and gamer looking for a CPU upgrade, this Ryzen 9 9900X deal will get you $90 offAMD is now reportedly making all-American Ryzen 9000 CPU dies at TSMC’s Arizona fabI tried a sim racing rig that generates ‘wind’ at CES 2025 and it’s claimed to help keep VR nausea at bayFinal Fantasy 14 communities panic as it turns out change to blacklisting, meant to help reduce stalking, also lets players use mods to track their altsTikTok’s time in the US may be ticking to a close as it makes a last stand in the Supreme Court, though Trump now says he ‘opposes banning’ itMSI intros cheaper ‘back connect’ Project Zero Intel motherboards for super-clean PC builds you might actually be able to afford

Most PopularMarvel Rivals’ latest update quietly killed the game’s burgeoning mod sceneSorry everyone, Minecraft 2 is cancelled, and it might be our faultTeam Fortress 2 just had the single biggest trade in its history, with ‘the absolute best hat you can get’ going for around $40KIs the new RTX 5070 really as fast as Nvidia’s previous flagship RTX 4090 GPU? Turns out the answer is yes. Kinda.Elden Ring: Nightreign is getting a closed network test, and I’m totally not fuming because PC players aren’t invited to the party, or anythingIf you’re a content creator and gamer looking for a CPU upgrade, this Ryzen 9 9900X deal will get you $90 offAMD is now reportedly making all-American Ryzen 9000 CPU dies at TSMC’s Arizona fabI tried a sim racing rig that generates ‘wind’ at CES 2025 and it’s claimed to help keep VR nausea at bayFinal Fantasy 14 communities panic as it turns out change to blacklisting, meant to help reduce stalking, also lets players use mods to track their altsTikTok’s time in the US may be ticking to a close as it makes a last stand in the Supreme Court, though Trump now says he ‘opposes banning’ itMSI intros cheaper ‘back connect’ Project Zero Intel motherboards for super-clean PC builds you might actually be able to afford

Most PopularMarvel Rivals' latest update quietly killed the game’s burgeoning mod sceneSorry everyone, Minecraft 2 is cancelled, and it might be our faultTeam Fortress 2 just had the single biggest trade in its history, with ‘the absolute best hat you can get’ going for around $40KIs the new RTX 5070 really as fast as Nvidia’s previous flagship RTX 4090 GPU? Turns out the answer is yes. Kinda.Elden Ring: Nightreign is getting a closed network test, and I’m totally not fuming because PC players aren’t invited to the party, or anythingIf you’re a content creator and gamer looking for a CPU upgrade, this Ryzen 9 9900X deal will get you $90 offAMD is now reportedly making all-American Ryzen 9000 CPU dies at TSMC’s Arizona fabI tried a sim racing rig that generates ‘wind’ at CES 2025 and it’s claimed to help keep VR nausea at bayFinal Fantasy 14 communities panic as it turns out change to blacklisting, meant to help reduce stalking, also lets players use mods to track their altsTikTok’s time in the US may be ticking to a close as it makes a last stand in the Supreme Court, though Trump now says he ‘opposes banning’ itMSI intros cheaper ‘back connect’ Project Zero Intel motherboards for super-clean PC builds you might actually be able to afford

Most Popular

Marvel Rivals' latest update quietly killed the game’s burgeoning mod scene

Sorry everyone, Minecraft 2 is cancelled, and it might be our fault

Team Fortress 2 just had the single biggest trade in its history, with ‘the absolute best hat you can get’ going for around $40K

Is the new RTX 5070 really as fast as Nvidia’s previous flagship RTX 4090 GPU? Turns out the answer is yes. Kinda.

Elden Ring: Nightreign is getting a closed network test, and I’m totally not fuming because PC players aren’t invited to the party, or anything

If you’re a content creator and gamer looking for a CPU upgrade, this Ryzen 9 9900X deal will get you $90 off

AMD is now reportedly making all-American Ryzen 9000 CPU dies at TSMC’s Arizona fab

I tried a sim racing rig that generates ‘wind’ at CES 2025 and it’s claimed to help keep VR nausea at bay

Final Fantasy 14 communities panic as it turns out change to blacklisting, meant to help reduce stalking, also lets players use mods to track their alts

TikTok’s time in the US may be ticking to a close as it makes a last stand in the Supreme Court, though Trump now says he ‘opposes banning’ it

MSI intros cheaper ‘back connect’ Project Zero Intel motherboards for super-clean PC builds you might actually be able to afford

HARDWARE BUYING GUIDESLATEST GAME REVIEWS1Best Steam Deck accessories in Australia for 2025: Our favorite docks, powerbanks and gamepads2Best graphics card for laptops: the mobile GPUs I’d want in my next gaming laptop3Best mini PCs in 2025: The compact computers I love the most4Best 14-inch gaming laptop: The top compact gaming laptops I’ve held in these hands5Best Mini-ITX motherboards in 2025: My pick from all the mini mobo marvels I’ve tested1Thank Goodness You’re Here! review: An anarchic treasure trove of jokes and skits2Shiren the Wanderer: The Mystery Dungeon of Serpentcoil Island review—like juggling chainsaws on horseback3WD Black SN850X 8 TB NVMe SSD review4Ikea Utespelare desk review5Asus ROG Harpe Ace Mini wireless mouse review

HARDWARE BUYING GUIDESLATEST GAME REVIEWS1Best Steam Deck accessories in Australia for 2025: Our favorite docks, powerbanks and gamepads2Best graphics card for laptops: the mobile GPUs I’d want in my next gaming laptop3Best mini PCs in 2025: The compact computers I love the most4Best 14-inch gaming laptop: The top compact gaming laptops I’ve held in these hands5Best Mini-ITX motherboards in 2025: My pick from all the mini mobo marvels I’ve tested1Thank Goodness You’re Here! review: An anarchic treasure trove of jokes and skits2Shiren the Wanderer: The Mystery Dungeon of Serpentcoil Island review—like juggling chainsaws on horseback3WD Black SN850X 8 TB NVMe SSD review4Ikea Utespelare desk review5Asus ROG Harpe Ace Mini wireless mouse review

HARDWARE BUYING GUIDESLATEST GAME REVIEWS1Best Steam Deck accessories in Australia for 2025: Our favorite docks, powerbanks and gamepads2Best graphics card for laptops: the mobile GPUs I’d want in my next gaming laptop3Best mini PCs in 2025: The compact computers I love the most4Best 14-inch gaming laptop: The top compact gaming laptops I’ve held in these hands5Best Mini-ITX motherboards in 2025: My pick from all the mini mobo marvels I’ve tested1Thank Goodness You’re Here! review: An anarchic treasure trove of jokes and skits2Shiren the Wanderer: The Mystery Dungeon of Serpentcoil Island review—like juggling chainsaws on horseback3WD Black SN850X 8 TB NVMe SSD review4Ikea Utespelare desk review5Asus ROG Harpe Ace Mini wireless mouse review

HARDWARE BUYING GUIDESLATEST GAME REVIEWS1Best Steam Deck accessories in Australia for 2025: Our favorite docks, powerbanks and gamepads2Best graphics card for laptops: the mobile GPUs I’d want in my next gaming laptop3Best mini PCs in 2025: The compact computers I love the most4Best 14-inch gaming laptop: The top compact gaming laptops I’ve held in these hands5Best Mini-ITX motherboards in 2025: My pick from all the mini mobo marvels I’ve tested1Thank Goodness You’re Here! review: An anarchic treasure trove of jokes and skits2Shiren the Wanderer: The Mystery Dungeon of Serpentcoil Island review—like juggling chainsaws on horseback3WD Black SN850X 8 TB NVMe SSD review4Ikea Utespelare desk review5Asus ROG Harpe Ace Mini wireless mouse review

HARDWARE BUYING GUIDESLATEST GAME REVIEWS

1Best Steam Deck accessories in Australia for 2025: Our favorite docks, powerbanks and gamepads

A bunch of the best Steam Deck accessories on a blue background.

1Best Steam Deck accessories in Australia for 2025: Our favorite docks, powerbanks and gamepads

1

Best Steam Deck accessories in Australia for 2025: Our favorite docks, powerbanks and gamepads

2Best graphics card for laptops: the mobile GPUs I’d want in my next gaming laptop

2Best graphics card for laptops: the mobile GPUs I’d want in my next gaming laptop

2

Best graphics card for laptops: the mobile GPUs I’d want in my next gaming laptop

3Best mini PCs in 2025: The compact computers I love the most

3Best mini PCs in 2025: The compact computers I love the most

3

Best mini PCs in 2025: The compact computers I love the most

4Best 14-inch gaming laptop: The top compact gaming laptops I’ve held in these hands

4Best 14-inch gaming laptop: The top compact gaming laptops I’ve held in these hands

4

Best 14-inch gaming laptop: The top compact gaming laptops I’ve held in these hands

5Best Mini-ITX motherboards in 2025: My pick from all the mini mobo marvels I’ve tested

5Best Mini-ITX motherboards in 2025: My pick from all the mini mobo marvels I’ve tested

5

Best Mini-ITX motherboards in 2025: My pick from all the mini mobo marvels I’ve tested

1Thank Goodness You’re Here! review: An anarchic treasure trove of jokes and skits

Thank Goodness You’re Here review

1Thank Goodness You’re Here! review: An anarchic treasure trove of jokes and skits

1

Thank Goodness You’re Here! review: An anarchic treasure trove of jokes and skits

2Shiren the Wanderer: The Mystery Dungeon of Serpentcoil Island review—like juggling chainsaws on horseback

2Shiren the Wanderer: The Mystery Dungeon of Serpentcoil Island review—like juggling chainsaws on horseback

2

Shiren the Wanderer: The Mystery Dungeon of Serpentcoil Island review—like juggling chainsaws on horseback

3WD Black SN850X 8 TB NVMe SSD review

3WD Black SN850X 8 TB NVMe SSD review

3

WD Black SN850X 8 TB NVMe SSD review

4Ikea Utespelare desk review

4Ikea Utespelare desk review

4

Ikea Utespelare desk review

5Asus ROG Harpe Ace Mini wireless mouse review

5Asus ROG Harpe Ace Mini wireless mouse review

5

Asus ROG Harpe Ace Mini wireless mouse review