--- title: 'The AI did it' is not a legal defense, and never was description: When AI agents take real-world actions and cause harm, the humans and organizations deploying them bear full legal responsibility — not the AI, not the vendor.... canonical: https://daily.dev/posts/the-ai-did-it-is-not-a-legal-defense-and-never-was-c3aro1ywg twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: 'The AI did it' is not a legal defense, and never was | daily.dev og:description: When AI agents take real-world actions and cause harm, the humans and organizations deploying them bear full legal responsibility — not the AI, not the vendor.... og:url: https://daily.dev/posts/the-ai-did-it-is-not-a-legal-defense-and-never-was-c3aro1ywg og:image: https://api.daily.dev/og/posts/c3ARo1yWG.png og:image:alt: 'The AI did it' is not a legal defense, and never was og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # 'The AI did it' is not a legal defense, and never was **[Trends](https://daily.dev/sources/trends)** · 2 min read · 1 upvotes · 1 comments ## Summary When AI agents take real-world actions and cause harm, the humans and organizations deploying them bear full legal responsibility — not the AI, not the vendor. Approval-based workflows don't distribute blame; they concentrate it. Clicking through agent confirmations without genuine understanding is legally indefensible. Teams building agentic systems need to treat human review as real oversight, not a liability shield. ## Content Half of AI-generated security patches don't actually fix the vulnerability. That's the headline from 1Password's FLAWED study, which ran 6,080 patches through ChatGPT-5.5 and Claude Opus 4.8 across six real CVEs. Only 26% fully remediated the flaw without changing application behavior. Another 20% fixed it by altering behavior in ways that may or may not be acceptable. The remaining 49%? Still exploitable. Worse, over a third of patches that *passed* initial testing were "fragile" — they blocked the specific proof-of-concept exploit without touching the root cause. Veracode's parallel research across 100+ models found 44% of AI-generated code introduces OWASP Top 10 vulnerabilities. So the models are better at finding holes than plugging them. That's a useful thing to know. The more interesting fight is over what happens next — specifically, who owns the mess when an AI patch ships and breaks something. The discourse is converging on a pretty uncomfortable answer: you do. @QuinnyPig put it bluntly: "If I fail to properly supervise my children at the museum, I'm about to have a bad time when asked to pay for the priceless artifact they just destroyed. Agents can't hold legal liability; companies and humans can." @copyconstruct adds a subtler point — things that look fine in isolation when an agent wrote a patch can look very different from a panoramic view. APIs get misshapen. Abstractions get too complex. "Maintenance is going to require constant gardening and pruning. Exhausting." The "meat proxy" framing is getting traction: the person who rubber-stamps AI output without actually understanding it isn't a reviewer, they're a liability with a keyboard. A Cursor study found 36% of AI-generated code changes are accepted automatically without manual review. That number should make you nervous. There's also a counterintuitive finding buried in the MIT research on medical AI: experts are *more* likely to miss AI mistakes than junior team members, because expertise runs on pattern recognition and AI output is optimized to match those patterns. The person who catches errors is usually whoever gets the 2am call when things break. @stolinski's take cuts through the noise: "Too many people use 'it's just one prompt to fix' as a crutch for sloppy work." The tools are genuinely useful. The oversight gap is the actual problem. ## Questions this post answers ### Are companies legally liable when an AI system they deployed makes a mistake? Yes. Organizations deploying AI systems cannot use 'the model did it' as a legal defense when things go wrong. Liability lands on the humans and companies who deployed and approved the system. AI agents cannot hold legal liability — only the people and organizations behind them can. _Teams shipping AI-powered products track how courts and regulators are framing this on daily.dev._ ### Why are senior engineers more likely to miss AI-generated errors than junior engineers? MIT research on medical AI found that experts are more likely to miss AI-generated mistakes than junior team members because expertise runs on pattern recognition — and AI output is specifically optimized to match those patterns. The person most likely to catch errors is whoever bears the direct consequences when things break, not the most experienced reviewer. _Engineers building AI review workflows find the research and debate on this on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 2 discussions and 18 comments across lobsters, hackernews (as of 2026-08-08). **TL;DR:** Commenters broadly agree with the spirit of the post — don't blindly relay AI output — but the discussion quickly branches into nuance: some defend forwarding AI responses in low-stakes situations, and others draw parallels to older "LMGTFY" culture. **Sentiment:** 40% positive · 45% mixed · 15% skeptical **The case for** - Verifying AI output before sharing it (e.g., asking the AI for a source, checking it, then sharing that) is a reasonable middle ground that adds genuine value. - LLMs are popular partly because they answer questions without the gatekeeping or dismissiveness that human experts sometimes show. **The pushback** - Some argue it's hypocritical to criticize AI-forwarding when the same community once weaponized "LMGTFY" links against questioners. - Not holding someone accountable for forwarding an AI response to a low-effort question is a defensible position — avoiding wasted time goes both ways. **By community** - lobsters (positive): Brief but warm reception, with a commenter drawing a fond parallel to nohello.com and another surfacing a related resource. - hackernews (mixed): Engaged but divided — commenters debate whether the criticism is consistent given past "LMGTFY" behavior, explore nuanced workarounds, and surface the central irony that people want AI-assisted creation but resent receiving it. **Hottest debate:** Whether criticizing meat-proxying is hypocritical given that the same communities once used dismissive "LMGTFY" links to deflect questions. **Open questions** - Where exactly is the line between acceptable AI-assisted responses and unacceptable meat-proxying, especially for routine or low-stakes questions? - Does the calculus change when the AI response is accurate and the questioner's time is genuinely saved? **Highlights** > It's becoming a central tension of our time: people want to create with AI but don't want to be downstream of anyone else creating with it. > — [gdulli on hackernews · 1 comments](https://news.ycombinator.com/item?id=49176840) > Yep. That said I hesitate to paste AI verbatim. Instead I tell the AI to give me a good source for the information, make sure it's trustworthy, skim it myself then paste that. Like how everyone used Wikipedia when schools didn't allow it as a source. > — [ragequittah on hackernews](https://news.ycombinator.com/item?id=49173435) > A definition of 'slop' I've come to like, although not perfect, is when something takes more effort to consume than to produce. > — [elianaive on hackernews · 1 comments](https://news.ycombinator.com/item?id=49174006) > LLMs are ascendant partially because they do not "RTFM", "read the FAQ", or "question closed as duplicate" at you regardless of how many times they've heard a question before. > — [Ariarule on hackernews](https://news.ycombinator.com/item?id=49176553) **Source threads** - [lobsters](https://lobste.rs/s/jtgaol/no_meat_proxy) · 20 points · 2 comments - [hackernews](https://news.ycombinator.com/item?id=49168157) · 27 points · 16 comments ## Community discussion Top comments from developers on daily.dev. **@petermrozek** · 0 upvotes > And did anybody really expect anything else? Companies (and management specifically) are always legally bound, so of course they need someONE to take responsilibily. That's why I call BS on the "fully autonomous agents" - if you want hallucinating automation, you'll have to take responsibility for it. Management will always conveniently point at the one who "pressed the button". 😉 > > And AI companies? They have a fine print in their ToS saying the technology is experimental, can hallucinate and you use it at your own risk... while simultaniously charging you for using them. 😘 ## Similar posts on daily.dev - [AI can find zero-days but still can’t reliably write secure code](https://daily.dev/posts/ai-can-find-zero-days-but-still-can-t-reliably-write-secure-code-gkwz7v79e) · CSO Online · 0 upvotes · 0 comments - [AI gets good at finding bugs, not as good at fixing them](https://daily.dev/posts/ai-gets-good-at-finding-bugs-not-as-good-at-fixing-them-sptd5ofrp) · The Register · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents) [View this post on daily.dev](https://daily.dev/posts/the-ai-did-it-is-not-a-legal-defense-and-never-was-c3aro1ywg) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"'The AI did it' is not a legal defense, and never was","url":"https://daily.dev/posts/the-ai-did-it-is-not-a-legal-defense-and-never-was-c3aro1ywg","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/the-ai-did-it-is-not-a-legal-defense-and-never-was-c3aro1ywg"},"datePublished":"2026-08-01T06:36:19.939Z","dateModified":"2026-08-08T02:47:47.489Z","description":"When AI agents take real-world actions and cause harm, the humans and organizations deploying them bear full legal responsibility — not the AI, not the vendor....","image":"https://i.ytimg.com/vi/gSeadaokozg/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/gSeadaokozg/sddefault.jpg","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/the-ai-did-it-is-not-a-legal-defense-and-never-was-c3aro1ywg","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"ai-agents","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"'The AI did it' is not a legal defense, and never was"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/the-ai-did-it-is-not-a-legal-defense-and-never-was-c3aro1ywg","comment":[{"@type":"Comment","text":"And did anybody really expect anything else? Companies (and management specifically) are always legally bound, so of course they need someONE to take responsilibily. That’s why I call BS on the “fully autonomous agents” - if you want hallucinating automation, you’ll have to take responsibility for it. Management will always conveniently point at the one who “pressed the button”. 😉\nAnd AI companies? They have a fine print in their ToS saying the technology is experimental, can hallucinate and you use it at your own risk… while simultaniously charging you for using them. 😘","datePublished":"2026-08-02T06:48:44.112Z","dateModified":"2026-08-02T06:52:28.953Z","url":"https://daily.dev/posts/c3ARo1yWG#c-Uy86NFSaT","author":{"@type":"Person","name":"Peter Mrożek","url":"https://daily.dev/petermrozek","image":"https://media.daily.dev/image/upload/s--pBfYX68K--/f_auto/v1769247960/avatars/avatar_Qz65P1nVw3Bu6C5YwaZgA?_a=BAMAMiiu0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/the-ai-did-it-is-not-a-legal-defense-and-never-was-c3aro1ywg#faq","mainEntity":[{"@type":"Question","name":"Are companies legally liable when an AI system they deployed makes a mistake?","acceptedAnswer":{"@type":"Answer","text":"Yes. Organizations deploying AI systems cannot use 'the model did it' as a legal defense when things go wrong. Liability lands on the humans and companies who deployed and approved the system. AI agents cannot hold legal liability — only the people and organizations behind them can. Teams shipping AI-powered products track how courts and regulators are framing this on daily.dev."}},{"@type":"Question","name":"Why are senior engineers more likely to miss AI-generated errors than junior engineers?","acceptedAnswer":{"@type":"Answer","text":"MIT research on medical AI found that experts are more likely to miss AI-generated mistakes than junior team members because expertise runs on pattern recognition — and AI output is specifically optimized to match those patterns. The person most likely to catch errors is whoever bears the direct consequences when things break, not the most experienced reviewer. Engineers building AI review workflows find the research and debate on this on daily.dev."}}]} ```