[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-llm-safety-guardrails-have-a-bypass-built-in":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},3173,"llm-safety-guardrails-have-a-bypass-built-in","LLM Safety Guardrails Have a Bypass Built In","Researchers found that LLM safety guardrails sit on a linear 'refusal axis' that can be flipped in about one second, or hardened without retraining.","LLM safety guardrails are more like a dial than a wall.\n\nResearchers studied the \"refusal direction\" embedded in large language models by contrasting how models respond to safe versus unrestricted system prompts. Their method, called Contrastive Logit Steering (CLS), works on the model's output distribution rather than its internal activations — and when paired with a simple prefix injection, it triggers what the paper calls a phase transition where safety guardrails collapse entirely. Testing across seven model families, the team found that safety architecture is deterministic: Llama-3.1 models use a \"Late Decision\" pattern that CLS bypasses at a 95% attack success rate in roughly one second, while Qwen-2.5 integrates safety earlier in computation, making it harder to flip. CLS also outperformed existing hidden-state steering methods, hitting 73% success against Llama 2 where prior techniques reached 22.6%.\n\nThe same linearity that makes guardrails breakable also makes them tunable: inverting the steering vector can harden a model against jailbreaks without any retraining. That's the part that matters for defenders — a lightweight, no-training patch could be applied post-deployment. But the attack-side findings reinforce what critics of behavioral alignment have long argued: if safety is just a direction in latent space, it's a policy, not a principle.\n\nThe broader implication is that alignment techniques relying on output-layer steering may be more fragile than their benchmark scores suggest — and that the gap between \"safe-by-training\" and \"safe-by-architecture\" is still very wide.","[\"ai\",\"security\",\"alignment\",\"jailbreak\"]","2026-07-01T04:00:00.000Z","2026-07-01T08:58:55.722Z","2026-07-01T08:58:58.523Z","published",null,[],"ai",[24,26,27,28],"security","alignment","jailbreak",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.22686",0,{"sections":35},[36,40,44,49,54,59,64,69,74,79,84,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":26,"count":42,"latest_published_at":43},"Security",294,"2026-07-15T19:59:48.000Z",{"name":45,"slug":46,"count":47,"latest_published_at":48},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":50,"slug":51,"count":52,"latest_published_at":53},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":55,"slug":56,"count":57,"latest_published_at":58},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":60,"slug":61,"count":62,"latest_published_at":63},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":65,"slug":66,"count":67,"latest_published_at":68},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":75,"slug":76,"count":77,"latest_published_at":78},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":80,"slug":81,"count":82,"latest_published_at":83},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":85,"slug":86,"count":82,"latest_published_at":87},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]