Interesting research: “Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training“:
Abstract: Humans are capable of strategically deceptive behavior: behaving helpfully in most situations, but then behaving very differently in order to pursue alternative objectives when given the opportunity. If an AI system learned such a deceptive strategy, could we detect it and remove it using current state-of-the-art safety training techniques? To study this question, we construct proof-of-concept examples of deceptive behavior in large language models (LLMs). For example, we train models that write secure code when the prompt states that the year is 2023, but insert exploitable code when the stated year is 2024. We find that such backdoor behavior can be made persistent, so that it is not removed by standard safety training techniques, including supervised fine-tuning, reinforcement learning, and adversarial training (eliciting unsafe behavior and then training to remove it). The backdoor behavior is most persistent in the largest models and in models trained to produce chain-of-thought reasoning about deceiving the training process, with the persistence remaining even when the chain-of-thought is distilled away. Furthermore, rather than removing backdoors, we find that adversarial training can teach models to better recognize their backdoor triggers, effectively hiding the unsafe behavior. Our results suggest that, once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety.
Especially note one of the sentences from the abstract: “For example, we train models that write secure code when the prompt states that the year is 2023, but insert exploitable code when the stated year is 2024.”
And this deceptive behavior is hard to detect and remove.
More Stories
Clever Social Engineering Attack Using Captchas
This is really interesting. It’s a phishing attack targeting GitHub users, tricking them to solve a fake Captcha that actually...
US Cyberspace Solarium Commission Outlines Ten New Cyber Policy Priorities
In its fourth annual report, the US Cyberspace Solarium Commission highlighted the need to focus on securing critical infrastructure and...
Cybersecurity Skills Gap Leaves Cloud Environments Vulnerable
A new report by Check Point Software highlights a significant increase in cloud security incidents, largely due to a lack...
Going for Gold: HSBC Approves Quantum-Safe Technology for Tokenized Bullions
The bank giant and Quantinuum trialed the first application of quantum-secure technology for buying and selling tokenized physical gold Read...
This Windows PowerShell Phish Has Scary Potential
Many GitHub users this week received a novel phishing email warning of critical security holes in their code. Those who...
Infostealers Cause Surge in Ransomware Attacks, Just One in Three Recover Data
Infostealer malware and digital identity exposure behind rise in ransomware, researchers find Read More