Brendan Murphy

Dillon Bowen, Brendan Murphy, Will Cai et al.

LLMs produce harmful and undesirable behavior when trained on datasets containing even a small fraction of poisoned data. We demonstrate that GPT models remain vulnerable to fine-tuning on poisoned data, even when safeguarded by moderation systems. Given the persistence of data poisoning vulnerabilities in today's most capable models, this paper investigates whether these risks increase with model scaling. We evaluate three threat models -- malicious fine-tuning, imperfect data curation, and intentional data contamination -- across 24 frontier LLMs ranging from 1.5 to 72 billion parameters. Our experiments reveal that larger LLMs are significantly more susceptible to data poisoning, learning harmful behaviors from even minimal exposure to harmful data more quickly than smaller models. These findings underscore the need for leading AI companies to thoroughly red team fine-tuning APIs before public release and to develop more robust safeguards against data poisoning, particularly as models continue to scale in size and capability.

Nusrat Zahan, Thomas Zimmermann, Patrice Godefroid et al.

Modern software development frequently uses third-party packages, raising the concern of supply chain security attacks. Many attackers target popular package managers, like npm, and their users with supply chain attacks. In 2021 there was a 650% year-on-year growth in security attacks by exploiting Open Source Software's supply chain. Proactive approaches are needed to predict package vulnerability to high-risk supply chain attacks. The goal of this work is to help software developers and security specialists in measuring npm supply chain weak link signals to prevent future supply chain attacks by empirically studying npm package metadata. In this paper, we analyzed the metadata of 1.63 million JavaScript npm packages. We propose six signals of security weaknesses in a software supply chain, such as the presence of install scripts, maintainer accounts associated with an expired email domain, and inactive packages with inactive maintainers. One of our case studies identified 11 malicious packages from the install scripts signal. We also found 2,818 maintainer email addresses associated with expired domains, allowing an attacker to hijack 8,494 packages by taking over the npm accounts. We obtained feedback on our weak link signals through a survey responded to by 470 npm package developers. The majority of the developers supported three out of our six proposed weak link signals. The developers also indicated that they would want to be notified about weak links signals before using third-party packages. Additionally, we discussed eight new signals suggested by package developers.

Marcus Williams, Micah Carroll, Adhyyan Narang et al. · berkeley

As LLMs become more widely deployed, there is increasing interest in directly optimizing for feedback from end users (e.g. thumbs up) in addition to feedback from paid annotators. However, training to maximize human feedback creates a perverse incentive structure for the AI to resort to manipulative or deceptive tactics to obtain positive feedback from users who are vulnerable to such strategies. We study this phenomenon by training LLMs with Reinforcement Learning with simulated user feedback in environments of practical LLM usage. In our settings, we find that: 1) Extreme forms of "feedback gaming" such as manipulation and deception are learned reliably; 2) Even if only 2% of users are vulnerable to manipulative strategies, LLMs learn to identify and target them while behaving appropriately with other users, making such behaviors harder to detect; 3) To mitigate this issue, it may seem promising to leverage continued safety training or LLM-as-judges during training to filter problematic outputs. Instead, we found that while such approaches help in some of our settings, they backfire in others, sometimes even leading to subtler manipulative behaviors. We hope our results can serve as a case study which highlights the risks of using gameable feedback sources -- such as user feedback -- as a target for RL.

Brendan Murphy, Dillon Bowen, Shahrad Mohammadzadeh et al.

AI systems are rapidly advancing in capability, and frontier model developers broadly acknowledge the need for safeguards against serious misuse. However, this paper demonstrates that fine-tuning, whether via open weights or closed fine-tuning APIs, can produce helpful-only models with safeguards destroyed. In contrast to prior work which is blocked by modern moderation systems or achieved only partial removal of safeguards or degraded output quality, our jailbreak-tuning method teaches models to generate detailed, high-quality responses to arbitrary harmful requests. For example, OpenAI, Google, and Anthropic models will fully comply with requests for CBRN assistance, executing cyberattacks, and other criminal activity. We further show that backdoors can increase not only the stealth but also the severity of attacks. Stronger jailbreak prompts become even more effective in fine-tuning attacks, linking attacks and potentially defenses in the input and weight spaces. Not only are current models vulnerable, more recent ones also appear to be becoming even more vulnerable to these attacks, underscoring the urgent need for tamper-resistant safeguards. Until such safeguards are discovered, companies and policymakers should view the release of any fine-tunable model as simultaneously releasing its evil twin: equally capable as the original model, and usable for any malicious purpose within its capabilities.