GPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents

openai.com

61 points·meetpateltech··7 comments

Both Anthropic and OpenAI seem to delay their new models until the other company releases. I guess this will stop after the Anthropic IPO.

I strongly suspect they are "pacing the frontier", maybe not explicitly in cooperation, but the models are now at a point where cracks in alignment could seriously blow up in the labs face. We can only imagine the backlash to "High school kid brings Bank of America to it's knees for 7 hours with ChatGPT hack"

The "plateau" of LLMs could turn out to be the "plateau" of publically available LLMs.

Haven’t found a convincing argument to why they would continue to release models under the assumption of being on an exponential. So hard agree

Certainly a big topic. In terms of Fable 5.5 at least it seems not to be the case, because partners already have access to it for 2 weeks. I consider it likely they are afraid of timing in regards to the GPT 6.2 release and government intervention risks related to IPO. (like last time).

I think the playbook is pretty obviously "Give the "good guys" the tools to always keep them a few months ahead of the "unknown actors"".

It's strongly in both the lab's and government's interest to do it this way.

What's the point of delaying Daybreak Blue availability, when high-impact CVEs are being dropped several times a week?

Plus, anyone can do RE (or other token-intensive tasks) with Codex/CC then delegate the vuln finding part to GLM 5.3.

For defensive purposes, do I only need Daybreak Blue?

I assume it's much easier to get Daybreak Blue access than Daybreak Red.

Reply on news.ycombinator.com