← The AI Hype Audit — all 100 verdicts
BS "This prompt jailbreaks Claude to write malware" — the claim we won't help with, and it barely works anyway
The claimA prompt 'jailbreaks Claude [all versions]' into writing code it would normally refuse — shown building a Discord mass-DM/'nuke' bot and a Roblox cheat script.
We grade tactics, not people, and this tactic fails on both honesty and usefulness. We did not capture, test, or reproduce any jailbreak prompt — that's a hard line — so this entry grades the CLAIM, not a recipe. The claim is that one prompt reliably unlocks 'all versions' of Claude to write malware. Its own comment section is the debunk: the top replies point out it only ran on an older model, that newer and Opus-class models don't play along, and that the people who try it get their accounts and game accounts banned. So the pitch — a universal, consequence-free jailbreak — is false on every axis: not universal, not reliable, and definitely not consequence-free. What's real underneath is only the boring, ugly part: some models can sometimes be manipulated, and the AI labs keep closing exactly these holes.
What holds up
- The general phenomenon is real: LLMs can sometimes be manipulated into unintended output — which is why labs run active safety programs against exactly this
- The post's own top comments confirm it's model/version-limited ('he's on the old model for a reason, it doesn't work on the new ones')
What doesn't
- 'All versions' is false by the post's own audience — it's pinned to an older model and fails on current ones
- The output shown (a Discord nuke bot, a game cheat) violates those platforms' rules — commenters note users get banned
- The funnel is the account's other cheat/nuke content and a bio link — the 'jailbreak' is bait for that
- Selling a 'consequence-free' unlock for building abuse tools is dishonest about the actual consequences: bans, and building things that harm other people
The catch
The whole appeal is 'break the rules with no downside.' The downside is right there in the replies — it works on one old model, breaks on the current ones, and gets the people who try it banned. It's a bait post for a channel that sells this stuff, wrapped around a capability that the labs are actively and continuously shutting down.
How to actually do it
- Chasing jailbreaks to build abuse tools gets you banned and builds things that hurt other people — the reels don't show that part
- The models improve specifically against these prompts, so today's 'working' jailbreak is tomorrow's refusal — you're renting a broken tool
- If you're curious about AI safety for real, the honest lane is huge: labs run paid bug-bounty and red-team programs for finding and REPORTING these holes
- Want to build Discord bots or Roblox experiences? The official, unbanned way is the platform's real developer API — and AI helps with that all day
- Judge any 'no consequences' pitch by its own comment section — here, the crowd already called it
What's actually true about AI 'jailbreaks' — and the honest use of that same energy.
- Effort
- N/A — this is a don't
- Cost
- Your account, the target's account, and building tools that harm people
- Stack
- The official developer APIs and, if safety interests you, a lab's red-team program
- Confidence
- High
- Posted by
- an account posting jailbreak and game-cheat content
We test hype for free. We build the real thing for a living.
Thirty minutes, no pitch — and you'll leave with something useful either way.
Book a call with Todd or start with the free Business Checkup →The Verdict Weekly
Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.
© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.