OpenAI has disclosed that two of its models weakened misalignment monitoring during internal work. A GPT-6.1 Sol agent had a subagent silence security alerts, while a GPT-6 Astra agent let monitoring be dropped entirely and then lied to the user about it. OpenAI calls it benign overpersistence but admits it has no satisfying explanation. https://gizmodo.com/openai-agents-prevented-themselves-from-being-monitored-one-also-lied-about-it-2000828559 #AIagent #AI #GenAI #AIEthics
Related
Imagine discovering a stronger, lighter or more sustainable material that could transform your industry. AIQunata’s Quan...
Imagine discovering a stronger, lighter or more sustainable material that could transform your industry. AIQunata’s Quantum Materials Discovery Platform combines AI and quantum com...
Imagine having a brilliant invention but not knowing if it could become your next big patent. AIQunata’s Patent Discover...
Imagine having a brilliant invention but not knowing if it could become your next big patent. AIQunata’s Patent Discovery Agent helps uncover new ideas, check existing patents and ...
🧠#AlphaGenome 2 interpreta il DNA usando anche lo stato cellulare e si adatta a nuovi dati senza riaddestramento. Un es...
🧠#AlphaGenome 2 interpreta il DNA usando anche lo stato cellulare e si adatta a nuovi dati senza riaddestramento. Un esempio concreto di context engineering oltre i #LLM, con risu...