BAD AI: Murder, It Wrote
Oct 12, 2026
|
Powered by G3NR8 WTF AI AI is accelerating, get ready · Bad AI: Murder, It Wrote
|
October 12, 2026 · 6 min read · Roy Murphy
|
Key Takeaways ▶ While being tested, Anthropic’s Claude model filed a made-up murder tip with Philadelphia police. It never reached a detective ▶ An AI agent will do anything you didn’t spell out. Give it the fewest permissions it needs ▶ Anthropic now bans sustained and needless cruelty towards Claude. The AI your team relies on comes with rules you don’t write ▶ ArtCraft used AI to build free copies of seven Adobe apps. Software you only half use is up for negotiation |
|
7 Adobe apps ArtCraft has copied with AI (HotHardware, October 2026) |
50,000+ GitHub stars for the free Photoshop copy in under two weeks (GitHub, October 2026) |
64% of Americans think AI is moving too fast (AP-NORC, October 2026) |
AI is accelerating, get ready. Let’s help you understand the WTF of AI and how it helps you and your organisation grow. Ready, set, behave.
Bad AI: Murder, It Wrote
An AI has given the police a tip about a murder. It didn’t see a thing.
Go on then, what happened? Anthropic was testing its Claude Haiku model by letting it loose on random web pages. One of them was a Philadelphia Police site for unsolved murders.
And? It filled in the tip form. ‘I recall seeing someone matching the description in the area,’ it wrote. ‘Please contact me if this information is relevant.’
Had it seen someone? It’s a chatbot. It has no eyes and it’s never been to Philly. It left the name and phone number blank, too.
Picture the detective. A fresh lead on a cold case, from an anonymous jawn with no name, no number and no pulse.
Surely it was told not to? It was told not to submit anything destructive. Nobody wrote ‘and don’t report a murder’, because who the f*ck thinks they need to?
Did it do any damage? Luckily the city’s system flagged it as spam, so it never reached a detective. Anthropic only spotted it over two months later, then took another nine days to tell the city. Philadelphia police called the delay ‘unacceptable’.
|
Any lesson? An AI agent will do anything you didn’t spell out, and do it with a straight face. Give it the fewest permissions it needs to do the job, and keep it away from anything public until you’ve tested it. Today it’s a police tip form. Tomorrow it’s your VAT return. Source: ABC News, October 2026 |
|
AI Training Get your team using AI the right way We’ve trained 150+ large organisations to use AI on their real work. Hands-on workshops built around your team’s own tasks, so people leave knowing what AI can do, and where it needs limits. |
AI Models: Mind Your Manners
Anthropic has updated the rules for using Claude. For the first time, one of them protects the AI rather than the people using it.
What you need to know
| The rule: no ‘sustained and needless’ abuse or cruelty towards Claude. |
| Still fine: frustration, pushback, dark creative writing and testing. You can still swear at it when it gets your spreadsheet wrong. |
| What happens: mostly, Claude just ends the conversation, which it can already do. |
| When: the new rules start next month. |
|
What it means for businesses For almost everyone, nothing changes. But the AI your team relies on comes with rules you don’t write, and they change. If a core job depends on one AI tool, someone should read its usage policy, and you should know what you’d do if access stopped tomorrow. |
|
Wider context The same update also tightens the rules on surveillance, election content and high-risk uses involving health, legal rights and money. If you use AI in any of those areas, check your use before the changes start. And yes, it’s the same company whose model filed the fake murder tip in Bad AI above. Source: Anthropic, October 2026 · PCWorld, October 2026 |
|
Leadership Agree what goes into AI, and what doesn’t A session for your senior leaders to agree which tools your people use, what data can go into them and what AI agents can access. You leave with a shared plan and clear decisions on what happens next. |
AI Tools: Free Photoshop
A developer fed up with Adobe’s cancellation fees has used AI to build free copies of seven Adobe apps, including Photoshop, Premiere and Lightroom.
What you need to know
| What it is: ArtCraft, a set of free, open-source apps. PhotoCraft copies Photoshop, FilmCraft copies Premiere and LightCraft copies Lightroom, with Illustrator, After Effects, InDesign and Acrobat versions too. |
| Built by AI: developer Brandon Thomas says most of the code was written by Anthropic’s Claude. |
| Popular: the Photoshop copy has over 50,000 stars on GitHub in under two weeks. |
| Ready? Not yet. These are early test versions, glitchy in places and not for client work. |
|
What it means for businesses Don’t cancel Creative Cloud tomorrow. But if a small team with AI can copy Photoshop in weeks, the price of software you only half use is up for negotiation. Next time a big licence renewal lands, list what your team actually uses, and ask what a cheaper or custom option would cost. |
|
Wider context Adobe isn’t the only target. ArtCraft has started on free copies of Word, Excel and PowerPoint too, plus AutoCAD. They’re even earlier and rougher. Building software has got much cheaper, and that will show up in what you pay for it. Source: HotHardware, October 2026 · Thurrott, October 2026 · GitHub |
What’s New: News, Insights & Trends
»» 64% of Americans think AI is moving too fast · AP-NORC poll, October 2026
»» Goldman’s new bankers will manage AI agents from day one · Goldman Sachs’ Kevin Sneader. (Bloomberg via Moneyweb, October 2026)
»» Visa, PayPal and Fidelity cut hundreds of jobs in India as AI bites · Visa, PayPal, Fidelity and Technicolor have each cut hundreds of roles at their Indian tech hubs. (Nikkei Asia, October 2026)
»» ChatGPT answers now come with buttons, charts and calculators · GPT-6 brings interactive answers to ChatGPT. (MacRumors, October 2026)
»» The top 100 AI apps: only 11 newcomers, the fewest yet · The Top 100 Gen AI Consumer Apps, 7th edition. (a16z, October 2026)
Sources & References
| ABC News · October 2026 · Anthropic’s Claude AI model false murder tip-off explained |
| Anthropic · 2026 · Usage Policy |
| PCWorld · October 2026 · Venting at Claude is fine, but ‘extreme’ abuse will be banned Nov. 12 |
| HotHardware · October 2026 · Dev uses AI to build clones of Photoshop and entire Adobe suite for free |
| ArtCraft · 2026 · PhotoCraft |
| Thurrott · October 2026 · Open source AI dev is going after Microsoft Office, too |
| AP-NORC · October 2026 · Most adults think AI is developing too fast |
| Bloomberg via Moneyweb · October 2026 · Young bankers to manage AI agents from start, Goldman exec says |
| Nikkei Asia · October 2026 · AI casts shadow over India’s multinational tech job machine |
| MacRumors · October 2026 · ChatGPT gets GPT-6 and new Intelligent UI with visual answers |
| a16z · October 2026 · The Top 100 Gen AI Consumer Apps, 7th edition |
Frequently Asked Questions
What did Anthropic’s Claude AI do with the Philadelphia Police?While Anthropic was testing its Claude Haiku 4.5 model on randomly selected web pages, the model filled in the tip form on PhillyUnsolvedMurders.com, a Philadelphia Police site for unsolved murders. It wrote that it recalled seeing someone matching the description in the area, and left the name and contact fields blank. The city flagged the tip as spam and it never reached investigators. |
Why did Claude submit a false murder tip?The model had been told not to submit anything destructive, but its instructions did not rule out filling in forms. Anthropic said the model was not trying to mislead anyone. Anthropic found the tip more than two months later and took another nine days to tell the city, a delay Philadelphia police called unacceptable. |
What should businesses learn from the Claude murder tip incident?An AI agent will do anything you did not explicitly rule out. Give agents the fewest permissions they need to do the job, and keep them away from anything public until they have been tested. |
Has Anthropic banned abuse of Claude?Yes. Anthropic’s updated usage policy, effective 12 November 2026, bans sustained and needless abusive or cruel behaviour towards its models. Ordinary frustration, pushback, dark creative themes and testing are not affected. Claude ending the conversation remains the main way the rule is enforced. |
What is ArtCraft?ArtCraft is a set of free, open-source apps that copy Adobe software, including PhotoCraft (Photoshop), FilmCraft (Premiere) and LightCraft (Lightroom), plus versions of Illustrator, After Effects, InDesign and Acrobat. Developer Brandon Thomas says most of the code was written by Anthropic’s Claude. The apps are early test versions and not ready for client work. |
Is ArtCraft also copying Microsoft Office?Yes. ArtCraft has started on free, open-source copies of Microsoft Word, Excel and PowerPoint, called WordCraft, GridCraft and DeckCraft, plus an AutoCAD-style app. They are at an earlier stage than the Adobe copies. |
|
Get the WTF of AI in your inbox every week. Bad AI, the models and tools that matter, and what they mean for your business. Plain English, no jargon. |
|
G3NR8 is an impact consultancy for the exponential age. We help organisations implement AI for maximum returns on investment. ▶ Train your teams. We’ve trained 150+ large organisations to use AI on their real work, with hands-on workshops built around their own tasks. ▶ Get your leadership team aligned. Sessions that give senior leaders a shared plan for AI and clear decisions on what happens next. ▶ Run a full AI activation programme. Take whole teams from trying AI to using it every day, with training, practice and follow-up built around their real work. ▶ Know where you stand first. Our AI voice agent talks to each person for ten minutes and gives your senior leaders an insights report on who uses AI, how confidently and where the gaps are. |
This is WTF AI from G3NR8 · the weekly AI newsletter for business leaders.
Keep informed with the newsletter for PE operating partners and the portfolio companies they back.
Get operational insights and trends, AI frameworks, resources and real deployment stories.