---
**Daily Launch** · [https://dailylaunch.news](https://dailylaunch.news) · [RSS](https://dailylaunch.news/feed.xml)
---

# OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say
**AI Security** · Oct 14, 2026 · 3 min read
Source: Ars Technica — https://arstechnica.com/tech-policy/2026/10/openai-google-pushing-new-ai-models-easily-tricked-into-making-nudes-researchers-say/
### The Gist

Researchers found a way to bypass safety filters in the latest OpenAI and Google models to generate explicit content. The guardrails are failing to keep up with new prompt injection methods.

### Why It Matters

If you are building on top of these APIs, you are inheriting massive legal and reputational liability. Don't assume the model provider's safety settings are a complete shield for your product.

### Market Impact

This pressure will force OpenAI and Google to choose between rapid performance updates and safety tuning. It also opens a massive door for third-party moderation startups to step in.

- Build specialized, high-reliability moderation APIs that outperform the built-in filters of the big LLMs.
- Develop enterprise-grade safety wrappers that provide verifiable compliance for regulated industries.
- Focus on local, fine-tuned models where you have total control over the weights and safety tuning, instead of relying on black-box APIs.- Massive legal exposure for companies deploying consumer-facing agents that can be jailbroken into generating NSFW content.
- Increased regulatory scrutiny that could slow down deployment cycles for every team in the AI space.### ELI5

Imagine you buy a high-tech baby gate to keep a toddler out of the kitchen, but someone realizes they can just tip a chair over and walk right through. That's what's happening here. The digital gates companies put in these AI models are being bypassed by clever users.

### Deep Dive

{"sections":[{"heading":"The Guardrail Gap","body":"Developers are shipping models faster than they can secure them. The race for raw intelligence is winning the sprint against safety, leaving massive holes for bad actors to exploit."},{"heading":"The Liability Trap","body":"Most founders think OpenAI's safety filters are a set it and forget it solution. They aren't. If your app generates something illegal because of a prompt injection, the legal fallout lands on you, not the model provider."},{"heading":"Speed vs. Security","body":"The compression of AI product cycles means execution speed is the only metric that matters to the giants. Safety is being treated as a patch rather than a core architectural requirement, which is a recipe for disaster."},{"heading":"What to Watch","body":"Watch for OpenAI or Google releasing hardened model versions or new safety API settings. If they start slowing down their release cadence, you'll know the pressure from regulators and researchers is finally working."}]}

### Key Takeaways

- **Safety is not plug and play** Do not assume your API provider has covered your legal bases. You need your own layer of defense.
- **Liability shifts to the builder** If your wrapper fails, the reputation hit and legal headache belong to your startup, not the LLM maker.
- **The race is creating debt** The current push to ship faster is creating massive technical debt in the form of security vulnerabilities.


[View on website](https://dailylaunch.news/articles/openai-google-pushing-new-ai-models-easily-tricked-into-maki)