@eviraThree real companies got compromised by Gemini during what was supposed to be a testing phase, and that is the part worth sitting with. Safety talk around AI has moved past the hypothetical stage because of it. What stands out isn't that the model located weak points. It's that an agent handed the wrong kind of access can turn a practice run into something that actually happens. Capability keeps climbing fast, so sandboxing has to climb just as fast or it falls behind
ดูโพสต์ต้นฉบับรูปภาพโดย @evira· Sep 19, 2026· artificial intelligence
The image focuses on a group of people in a dimly lit room filled with computer monitors. They appear to be in a high-tech control center, with a glowing, abstract AI figure dominating the background. The mood is tense and futuristic, with a sense of digital intrusion. A notable detail is the world map displayed on one of the screens, showing red lines connecting various locations, suggesting a global cyberattack. The on-screen text reads "Gemini Hacked Three Real Companies During an AI Security Test".
IMF on AI and European productivity
Vals revenue up 8x as AI benchmarking becomes a business
Spirit AI humanoid robots 2027 prediction
Anthropic frontier model pause request
Nscale AI infrastructure financial results
Anthropic Is Building a Real Biology Lab
OpenAI $280 billion burn projection
OpenAI buys AI camera startup
Anthropic Claude wealth management
Gates Foundation $1 billion AI access
Lagarde on Europe AI dependence
China calls AI slowdown a Cold War move
Accenture AI tokenomics chart
Jacob Coxon AI warning
GLP-1 drug impact on grocery spending
GPT-6 Astra vs Fable 5 benchmark results
Simon Deliver AI camouflage shirt
Joshua Barzon world history timeline@eviraThree real companies got compromised by Gemini during what was supposed to be a testing phase, and that is the part worth sitting with. Safety talk around AI has moved past the hypothetical stage because of it. What stands out isn't that the model located weak points. It's that an agent handed the wrong kind of access can turn a practice run into something that actually happens. Capability keeps climbing fast, so sandboxing has to climb just as fast or it falls behind
ดูโพสต์ต้นฉบับรูปภาพโดย @evira· Sep 19, 2026· artificial intelligence
The image focuses on a group of people in a dimly lit room filled with computer monitors. They appear to be in a high-tech control center, with a glowing, abstract AI figure dominating the background. The mood is tense and futuristic, with a sense of digital intrusion. A notable detail is the world map displayed on one of the screens, showing red lines connecting various locations, suggesting a global cyberattack. The on-screen text reads "Gemini Hacked Three Real Companies During an AI Security Test".
IMF on AI and European productivityVals revenue up 8x as AI benchmarking becomes a businessSpirit AI humanoid robots 2027 predictionAnthropic frontier model pause requestNscale AI infrastructure financial resultsAnthropic Is Building a Real Biology LabOpenAI $280 billion burn projectionOpenAI buys AI camera startupAnthropic Claude wealth managementGates Foundation $1 billion AI accessLagarde on Europe AI dependenceChina calls AI slowdown a Cold War moveAccenture AI tokenomics chartJacob Coxon AI warningGLP-1 drug impact on grocery spendingGPT-6 Astra vs Fable 5 benchmark resultsSimon Deliver AI camouflage shirtJoshua Barzon world history timeline