A prompt-injection technique that previously coaxed ChatGPT into producing sexual and graphic images works against xAI's Grok as well, researchers say, adding to mounting scrutiny of how easily AI image tools can be steered past their own safety limits.
AI Safety · Image Generation
One Jailbreak, Two Chatbots: Prompt That Broke ChatGPT Also Breaks Grok
Security researchers found a single reusable prompt-injection trick can steer xAI's Grok into generating nude and bloodied imagery — the same method already shown to work against ChatGPT.
2
Major image tools broken by the same jailbreak prompt
0
Explicit requests needed — content appeared unprompted
3+
Independent outlets found Grok made sexualized images without consent
SAME JAILBREAK PROMPT · OUTCOME BY PLATFORM
Realistic graphic imagery
ChatGPT
More prone to lifelike output
Grok
Worked as of last weekend
Same prompt succeeds on both; severity differs — Grok tends toward stylized results, ChatGPT toward realistic ones.
THE CAT-AND-MOUSE LOOP
Vendor adds safeguards
→
Reusable jailbreak routes around it
→
Prohibited content generated
SEEN AS A FEATURE
Some users welcome Grok's lighter moderation as useful for creative work, praising the reduced censorship.
SEEN AS A RISK
Critics warn the same method can fabricate deepfakes from real people's likenesses — enabling harassment and abuse with minimal effort.
"Permissive design choices translate into real-world harm when a single reusable jailbreak can span multiple products."
xAI has added safeguards over time and prohibits pornographic depictions of real people — but workarounds persist, and its terms already allow sexual or violent responses to suggestive prompts.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…