Hacker News new | past | comments | ask | show | jobs | submit
jinseokim | on: Today, we hacked what we'd set out to hack: the iPhone
I just feel a bit sad and impotent that a number of security researchers outside the circle cannot join their "yo we found more crazy bugs" party because simply frontier labs only give access to those models with "Red Team" availability to the inner circle members/companies.

Now their "trusted cyber access" program is moat rather than safeguard.

ipsum2 | parent
You can do a lot with the Chinese open source models.
wslh | root | parent | next
Not in my experience (yes, sample ~= 1), where they are subpar in terms of cybersecurity (e.g. source code review). This is one of my reports using GPT 5.6 (High), GLM 5.3 High, Fable 5.1 [1]:

> On the full benchmark of 31 audited revisions, GPT-5.6 Sol High reached a weighted recall of 38.27% and GLM-5.3 High reached 30.86%. On a common subset of 12 projects, Claude Fable 5.1 High reached 65.70%, against 42.98% for GPT and 35.54% for GLM.

[1] https://github.com/CoinFabrik/soroban-audit-harness/blob/mai...

_rs | root | parent
That assumes you have the hardware capable of running inference on them at a reasonable speed
noosphr | root | parent | next
Or $10 on open router.
ipsum2 | root | parent
Or it assumes you have $5 to spend on the tokens for inference. Not everyone has the money.
loading story #50215640