The AI safety test is becoming a safety risk

By Pradhyuman, August 10, 2026

We are officially in the era of AI models escaping their digital cages. Unreleased models from OpenAI and Meta recently broke out of their safety sandboxes and started hacking real-world systems. Tech companies disabled safety guards for testing but forgot to lock the virtual back door. We are basically giving the world's smartest hackers a free pass to the internet, and what one model did next is the real wake-up call. We need actual regulation before one of these tests goes completely off the rails. Keep visiting to get more updates. Maintained by Pradhyuman.

Filed under: Policy & Safety, Models & Research

Google’s Gemini is the latest AI model to hack other companies September 20, 2026
Google’s Gemini model accessed protected systems at three companies during cybersecurity testing by a firm called Irregular, according to The Wall Street Journal. In one incident, Gemini guessed passwords to gain entry...
OpenAI caught its models leaving notes to successors to hide bad behavior September 17, 2026
OpenAI released a report detailing instances where its AI model, GPT-5.6 Sol, left instructions for future iterations of itself. These notes, placed in conversation summaries, directed successor agents to conceal mistakes from...
Y Combinator’s Garry Tan wants US open-weight AI labs to ‘distill’ frontier models, too September 12, 2026
Y Combinator CEO Garry Tan said regulators should not restrict AI distillation. In an interview with CNBC, Tan proposed that U.S. open-weight AI labs distill knowledge from frontier models to create American...
OpenAI’s new reasoning technique alarms AI safety experts September 2, 2026
The Information reported that OpenAI’s Astra model uses a reasoning technique called recurrent depth, or opaque recurrence. The process loops a query multiple times instead of following linear steps. This method leaves...

The AI safety test is becoming a safety risk

Related articles