Interested in all things related to Computer Science. Also on BSky @profw.bsky.social
University of Surrey
Joined November 2011
- Tweets26.1K
- Following833
- Followers31.5K
- Likes2.8K
Not looking good huntress.com/blog/vendor-cre… and appears to be affecting M365 via Connectwise. Supply chain at play again. connectwise.com/company/trus…
An interesting piece of analysis “overseen” by @matthew_d_green - I was fascinated to see confirmation of how well studied lattice based cryptography is when I often hear it is “young” and thinly examined.
I was curious about a question: how much cryptanalytic work have human beings done on various problems, and how much AI inference does that represent? So I asked Claude and Astra to measure. bitsofwork.org
Classic McEliece looks like it’s officially out fir future developments bsi.bund.de/SharedDocs/Downl…
For those privacy minded people, and particularly those who may talk about about sensitive info in their car, a reminder for Dutch security services aivd.nl/actueel/nieuws/2026/…
No AI model is jailbreak-proof. The UK's AI Security Institute has found a universal jailbreak in every system it's tested. The real questions: who's watching the door, and do defenders keep their access? profwoodward.substack.com/p/…
🤦♂️
#HMDFuse: Sophisticated #AI, #CSAM detection, can't be disabled or circumvented.
"A phone incompatible with pornography", they said.
Turn it sideways.
paul.reviews/harmblock-world…
#childSafety #harmblock #AI #HMD #xplora #SafeToNet @NCSC @CEOPEducation @IWFhotline
Alan Woodward retweeted
The bots and I made a little thing for my cryptography class: the one-way function transform lab.
Interactively write security proofs (reductions + analysis) and attacks, both correct and wrong, for a variety of natural function transforms.
web.eecs.umich.edu/~cpeikert…
Jan 2026 the best structural attack on Classic McEliece cost 2^298+. By Aug, distinguishing cost ~2^120. Now full key recovery is claimed at 2^94–2^102 (ePrint 2026/1984), with an argument that no realistic parameters fix it. Nothing deployed is decryptable but margins vanishing.
For those in a risk group please read this - it’s not gonna make headlines but it does actually matter.
Today, the NCSC alongside the @FBI and the Dutch AIVD, has published an advisory exposing spyware used by Iranian state actors to target dissidents, activists, and journalists. 1/2
👇👇👇👇👇 worst thing BBC ever did was take this Click off the TV. And whilst he was a great presenter, he knew his stuff! Where has this level of common sense gone.
Amodei says recursive self-improvement "started this summer". A 75-page survey posted 2 days earlier by researchers building RSI commercially found the only two direct tests of what’s needed for exponential growth came back negative. arXiv:2609.11873 They can’t both be right.
This is a quote I would support.
"It's fine to warn people. It's even better to warn people with a solution. And it's absolutely inappropriate to make things up."
#FightTheFUD
not a bad quote that..... #AI #FUD #Cyber #Security
Looks like CRIME attack against TLS is much more powerful than originally thought as demonstrated by @kennyog and his team at @ETH_en eprint.iacr.org/2026/1972.pd… Now this something I will worry about - I suggest cyber security folk read it too.
AI models can hack. What they can't do is judge: no doubt, no stakes, no knowing when to stop. Cyber security's police work stays human, as it always has been. Not a happy lot, but ours. I try to explain why here profwoodward.substack.com/p/…
Alan Woodward retweeted
We've reached the point where, after 25y in cybersecurity, I feel morally obligated to say this for the record:
The narrative being pushed around AI safety, sandbox incidents, and the suggestion that METR be treated as an authority is dangerous, deceptive, and morally corrupt.
Alan Woodward retweeted
How is Anthropic pushing regulatory capture? Here they promote METR, a "voluntary" third-party auditor that is not only ideologically and financially tied, but carries out their audits through unscientific means such as "CoT" and cherry-picks evidence to reinforce an idea of AGI.
