India
Joined December 2010
- Tweets4.8K
- Following174
- Followers252
- Likes10.8K
You should just put /goal and a mildly sensible prompt and have multiple agents working on their own goals.
It removes the hassle of back-n-forth responding to the agents and also provides you enough freedom to actually think and not just be an enabler of systems (yes/no) for your agents.
You can't even use the codex subscription for your own usecases now.
this is why we should have more frontier open-weight models
Codex on macOS: one long thread stalls on Thinking, then reconnects with "websocket closed by server before response.completed".
Forking it has the same issue; fresh chats work. Same model: fresh "hello" ~3s, fork ~116s to first message.
Anyone else seeing this?
Satyam retweeted
156 Indian government websites are currently inaccessible because of a cert revocation by @eMudhra, including 4 that come under @GoI_MeitY. One of them is @ccaindiapki - which regulates eMudhra.
cc @internetfreedom
I made a list: captnemo.in/blog/2026/10/23/…
The latest models refuse requests a lot now. I tried to replicate the emudhra situation locally and got flagged a bunch.
So much maneuvering to make non-exploitary experiments.
This will be one hell of an RCA report, the beginning of the agentic era postmortem format. Going to have a lot of learnings
Neoclouds will get resilient.
Folks have no idea how giants like AWS/GCP have had mammoth security issues pre-AI era and they came through with battle scars. All because of putting security foremost everything else.
I just read how Railway moved on from using the gcp managed KMS for storing the encryption keys to their own to lower down costs, and how it come to bite them now. But they are the only ones who have been transparent about how they operate in their engineering blogs. I found next to nothing on Fly or vercel
its time the industry shifted to epehemeral passkeys that auto-rotate at half-life instead of static api keys
An urgent notice from the Railway team
blog.railway.com/p/security-…
Satyam retweeted
For the next few days or weeks, anyone now has a root CA thanks to eMudhra. Time to have some fun with MiTMing unwilling devices!
emudhra has shared their response on the incident, and i have so many questions lol
> you used certified hsms and still ended up with 4 as a ca private key.
> two audits missed it for 18 months
> why even mention ai when someone could just try numbers below 256?
>how did your entire process let this through?
I work on production certificate-issuance infrastructure, and i'm struggling with the fact that one of e-mudhra's intermediate ca private keys was 4.
And the fact that folks just recovered it from the public certificate.
scotthelme.co.uk/eight-bits-…
p-384 requires an unpredictable choice from an enormous range. restricting the secret to one byte leaves 255 nonzero candidates. padding 4 to fill a 48-byte field changes its representation without adding randomness.