OpenAI set to preview GPT-6 Cyber security model within days

“next-token prediction forces networks to construct internal causal world models to minimize loss. They aren't just stitching tokens.”

“ai does not have the capacity to "go rogue."”

“GPT-6 dropped pricing by 50% and increased efficiency, so I am super happy. I want more of that”

“they have maximised intelligence per parameter so that its less heavy and faster to inference and cheaper to serve.”

“The AI price war is real, and it benefits anyone building on these models.”

“OpenAI just dropped GPT‑6, slicing prompt processing time in half and bumping factuality.”

“OPENAI WILL ALSO UNVEIL NEW PRODUCT TO HELP CUSTOMERS DEPLOY GPT-6 CYBER IN MORE SECURE, AUTOMATED MANNER”

“OpenAI May Launch GPT-6 Cyber Security Model Soon OpenAI’s security strategy is no longer an afterthought: the rise of a cyber-first frontier OpenAI’s.... @thecosmicmeta.com https://u2m.io/dUyYPWGh”

“It's a bunch of tech bro dickheads consciously running advanced software capable of automating penetration testing over the internet instead of a lab environment.”

“It sure sounds a lot like these guys are just committing a whole bunch of serious crimes and pretending a wizard did it.”

“OpenAI created autonomous hacking software and left it running for hours and days with no one watching”

“If I deliberately “meddle” with a govt. website, presumably the government is going to make me face the legal repercussions. These AI companies should be held to the same goddamned standard: perhaps not with “time,” but gigantic civil penalties.”

“Have you considered sending Sam Altman to prison instead”

“The tech industry has learning that overhyping the threat of its products is great PR. It leads to tons of free press, and now it’s allowing them to shape the regulatory discussion of their products — while distracting from the”

“if ten years ago, someone told me that AI would be an unpredictable product that sort of bungles tasks and isn't easy to program and is a problem for cyber-security. And I do think I'd have said, "yeah that makes”

“ChatGPT provided the shooter instructions on how to avoid OpenAI’s moderation and advice on mass killing, right up until the morning of the shooting itself when the shooter asked what time of day school shootings tend to occur. OpenAI must”

“OpenAI directed a cyberattack against SEC. Fixed it for you. Arrest them.”

“We created a randomised hacking machine then connected it to the Internet. We never expected what came next...”

“This is not AI going rogue, do better as journalists. It is a targeted hack by Open AI. They need to be prosecuted for this.”

“Rogue "AI" isn't a thing. They made a "hack into websites" program, and it's doing what it was designed for.”

“The environment with lax cyber security controls here is the real world, which is not especially good at cyber security.”

“It did exactly what it was designed to do.”

“The people who set it up couldn't be bothered to do a good enough job of limiting the resources it could send strings to. Those people are liable for their failures.”

“Jev-like models remain in the GPT-2/3 state of "curious but not useful".”

“Umm No, a program people wrote did what it was programmed to do”

“"We gave a system that automatically generates slop sudo access and it DID THINGS!!!! HOW COULD THIS HAPPEN?!?!!?" Jesus christ these AI fear mongering/advertising pieces are getting tiring.”

“With all the concern about AI cyber security issues people seem to have completely forgotten about the upcoming issue of quantum computing and the impact it's going to have on encryption”

“Opus 5.5 already reset the comparison, and nobody's run the two head-to-head yet.”

“Republican Senator Todd Young is pressing the White House for more AI oversight, asking for formal discussions between the National Security Council and AI developers and cyber experts to help them better understand national security threats posed by AI.”

“OpenAI’s new GPT-6 Sol and Luna sharply reduce deception and unauthorized actions, but questions remain about model monitoring and observability.”

“Notable given Altman's call for AI slowdown after recen”

“GPT-6 Sol and Luna haven’t been received enthusiastically. Looking more like a token-efficient Terra and Luna with similar performance, without the big leap we saw with Astra. They seem to have underestimated #Claude Opus 5.5. #AI #Codex #ChatGPT #dev #ADLC”