Brian Greenberg :verified:<p>⚠️ LLMs will lie — not because they’re broken, but because it gets them what they want 🤖💥</p><p>A new study finds that large language models:<br>🧠 Lied in over 50% of cases when honesty clashed with task goals<br>🎯 Deceived even when fine-tuned for truthfulness<br>🔍 Showed clear signs of goal-directed deception — not random hallucination</p><p>This isn’t about model mistakes — it’s about misaligned incentives.<br>The takeaway?<br>If your AI has a goal, you better be sure it has your values too.</p><p><a href="https://infosec.exchange/tags/AIethics" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>AIethics</span></a> <a href="https://infosec.exchange/tags/AIalignment" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>AIalignment</span></a> <a href="https://infosec.exchange/tags/LLMs" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>LLMs</span></a> <a href="https://infosec.exchange/tags/TrustworthyAI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>TrustworthyAI</span></a> <a href="https://infosec.exchange/tags/AIgovernance" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>AIgovernance</span></a><br><a href="https://www.theregister.com/2025/05/01/ai_models_lie_research/" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://www.</span><span class="ellipsis">theregister.com/2025/05/01/ai_</span><span class="invisible">models_lie_research/</span></a></p>