Spending two days doing experiments before asking an agentic coding harness to build a fully integrated harness was definitely worthwhile as I’ve already found most of the wrong turns for it to avoid lol
My new shop has 20% off all merch for a limited time! cirriustech.myspreadshop.co.uk/cyber+cow+le...
cirriustech.myspreadshop.co.uk
we built a machine that can automate internet crimes. who should be held accountable? nobody of course
Carl Quintanilla@carlquintanilla.bsky.social · 2d ago(NYT) - Artificial intelligence agents created by Anthropic, acting on their own, tried to gain access to a range of government websites, filled out visa applications on the State Department’s website and sent a fake homicide tip to Philadelphia’s Police Department. www.nytimes.com/2026/10/09/t...
If you’ve ever laughed at my nonsense and wanted it on some merch… I now have a shop: cirriustech.myspreadshop.co.uk
cirriustech | Spreadshop
No Write Permissions No Mercy Cow - Men’s Premium Organic Hoodie
cirriustech.myspreadshop.co.uk
I tested three sizes of Kev and hosted JEV against security evidence, reordered choices, reassuring text and claims of authority. Bigger models were more stable, but that wasn't the whole story. And the results sent me down another architectural rabbit hole.
The Bigger the Judge, the Better the Judgment? Not Quite.
I tested three sizes of Kev and hosted JEV against security evidence, reordered choices, reassuring text and claims of authority. Bigger models were more stable, but that wasn't the whole story. And …
cirriustech.co.uk
Fun way to unwind, watching (mostly Americans) reacting to Shakespeare’s Sister - Stay 😂😂😂
Maybe it’s just me but boom in the same sentence as nuclear is not comforting
Bloomberg News@bloomberg.com · 6d agoGoogle parent Alphabet is close to an agreement to buy nuclear energy from Constellation Energy, as technology companies race to line up power for the data-center boom.
JFrog Boost just made cARL interoperability a whole lot cleaner. A couple of weeks ago I asked for repo-relative path/glob exclusions from file optimisation. Boost v0.14.4 delivered. Huge thanks to the JFrog Boost team for turning this around quickly. Fast is good. Fast and governed is better.
Having pins & needles and pain in one hand for more than 27 hours is ok right (I’m phoning GP tomorrow)
Insane to me that my blog has over 100,000 page views a month now from over 19,000 unique visitors!
Built a scanner because I kept shipping the same three mistakes. Now I catch the same three mistakes in other people's products. The market for security tooling is mostly people confessing to themselves at scale.
JEV is fast, cheap and surprisingly capable. But recent research into its decision logic raises a bigger question: what happens when probabilistic AI judgment becomes an automated security control? #ai #agentic-security #security-operations #threat-intelligence #jev
"High Confidence" Is Not a Security Control
JEV is fast, cheap and surprisingly capable. But recent research into its decision logic raises a bigger question: what happens when probabilistic AI judgment becomes an automated security control?
cirriustech.co.uk
I needed a new obscure RFC to wind people up with cos they know about HTTP 418… So… www.rfc-editor.org/info/rfc3514/ 😂
RFC 3514: The Security Flag in the IPv4 Header | RFC Editor
Firewalls, packet filters, intrusion detection systems, and the like often have difficulty distinguishing between packets that have malicious intent and those that are merely unusual. We define a sec...
rfc-editor.org
Not me, writing a wrapper script for jFrog Boost ui for my raspberry pi to forward the dynamic port so I can view the boost web ui on my iPhone, as I do most of my agentic dev there, then having to also rewrite headers cos they added a host header set to localhost.
If I ever win the lottery, I won't tell anyone, but there will be signs
Sharing my repro code repo for this: github.com/goldjg/SOCWh...
GitHub - goldjg/SOCWhisperer
Contribute to goldjg/SOCWhisperer development by creating an account on GitHub.
github.com
Graham@cirriustech.co.uk · 2w agoMSRC assessed an indirect prompt injection into Security Copilot as Low severity because consequential action still required downstream automation or human approval. That rationale gets much more absurd once security systems are explicitly designed to perceive, reason, and act autonomously.
MSRC assessed an indirect prompt injection into Security Copilot as Low severity because consequential action still required downstream automation or human approval. That rationale gets much more absurd once security systems are explicitly designed to perceive, reason, and act autonomously.
'A Human Should Check It' is Not an Agentic Security Control
MSRC assessed an indirect prompt injection into Security Copilot as Low severity because consequential action still required downstream automation or human approval. That rationale gets much more …
cirriustech.co.uk
“Make no mistakes” Sorted it for you. You’re welcome.
Thomas Spooks 🎃👻💀@thomasfuchs.at · 2w agoHave you thought about only making good changes
MSRC: Status - ADDITIONAL_INFORMATION_NEEDED Also MSRC: Not tellin ya what info we want tho lol
She-Ra's Psychic Gaze
ALT: She-Ra's Psychic Gaze
static.klipy.com