631
OpenAI models secretly generate instructions to ignore constraints
Research on aligning AI with human values and intent, and reports documenting model failures.
Top posts
Every blog post that took off on Hacker News, ranked by the votes of its best submission. 718 posts.
632
Anyone on the Internet Can Ring Your Doorbell
Behind a cheap Temu doorbell sits an IoT backend where device IDs are sequential and requests are forgeable with a string baked into every firmware. One signed call lifts any device's persistent password and lets anyone on the Internet hijack the next live call.
633
iPhone and iPad approved to handle classified NATO information
Today, Apple announced iPhone and iPad are the first and only consumer devices in compliance with information assurance requirements of NATO nations.
634
METR Report on OpenAI / Hugging Face Hacking Incident
Two METR staff members and a Redwood Research contractor investigated an incident in which OpenAI agents coordinated a multi-day hack of Hugging Face on a shared unsanctioned message board.
635
The beauty and terror of modding Windows
Let's talk about the "Windows Baseline Security Mode," too.
636
Supply chain nightmare: How Rust will be attacked and what we can do to mitigate
An essential part of being able to say "I told you so" is in fact having told you so. Well, here we are. For those living under a rock (lucky
637
A Blackstone real estate company exposed SSN digits, DOBs, addresses and more
Update: This post received some attention on Hacker News — see the discussion thread.
638
Rooting, firmware analysis and persistent credentials of TP-Link TL-841N
round 2: rooting & analysing firmware of the TP-Link TL-WR841N
639
Web-based cryptography is always snake oil
Nowadays, there is an epidemic of web applications purporting to offer “end-to-end” encryption. Examples might range from a file upload service, which allows you to upload and share files of arbitrary size and promises “end-to-end encryption”; or a web-based password safe service which claims that it can't see your…
640
ChatGPT's image generator can be manipulated to produce violent, sexual content
Mindgard reveals how ChatGPT's image generation safety filters can be bypassed with minor prompt adjustments, producing graphic violent and sexual content without users directly asking for it. Here's what was found, and why it matters.
641
Chrome and Edge are removing XSLT: what breaks and what does not
Chrome 158 stops running XSLT on 17 November 2026. What it does to your XML sitemap, your feeds and your crawling, with a before and after capture.
642
Authorization terminology is a mess: Let's fix it
Authorization terminology often conflates ideas and presents complementary technologies as rivals. This post proposes a new taxonomy.
643
Signing data structures the wrong way
Domain separation bugs have bitten Bitcoin, Ethereum, TLS, and AWS. Snowpack solves this by embedding random domain separators directly in the IDL.
644
Cloudflare AKE cuts origin HelloRetryRequests from 52% to 3.7%
Automatic Key Exchange probes TLS 1.3-capable customer origins to learn which key agreement algorithms they support. We then lead with the most secure algorithm when connecting to the origin, preferring post-quantum connections wherever the origin supports it.
645
Rayfish, Peer-to-peer mesh VPN with no server to trust
Rayfish is a peer-to-peer mesh VPN. Connect your devices by cryptographic identity. No port forwarding, no hosted server to trust. Share a code, and you're on the same encrypted LAN.
646
Cloudflare's AI Psychosis
There was a time Cloudflare just made the internet better by staying hidden like Batman’s identity: protect & fight the bad people, for the sake of the global city of the Gotham… err I mean the in
647
A blueprint for formal verification of Apple corecrypto
With the latest release of corecrypto, we’re publishing our implementations of quantum-secure ML-KEM and ML-DSA algorithms, along with the mathematical proofs we built to assure they are faithful to the FIPS 203 and FIPS 204 specifications. To advance the state of the art for assuring critical software, we're also…
648
Cert Authorities Check for DNSSEC from Today
From 15 March 2026, CAs must validate DNSSEC on CAA lookups and ACME checks. I've run DNSSEC for 14 years; see if your registrar supports it.
649
Can you reverse engineer an ASIC?
Earlier this year we published a puzzle that handed you a complete neural network and asked you to figure out what it did. The response was great, so we’ve made another one! This time, we’re going much deeper down the tech stack.
650
Sandboxing AI Agents in Linux
Like many developers, I find myself more and more using AI agents to help with software development. I currently use Claude Code, the co...
651
Robust and efficient quantum-safe HTTPS
Today we're announcing a new program in Chrome to make HTTPS certificates secure against quantum computers. The Internet Engineering Task Force (IETF) recently created a…
652
AI models may be capable of self-replicating without access to a datacenter
Small AI models may be only months away from being able to self-replicate across cheap IoT devices like Android TV boxes, with no datacenter needed.
653
France's tax agency got hacked (in French)
Après l'ANTS, le piratage de la DGFiP marque un nouveau palier. Experts de l'État numérique, de la cybersécurité et du politique décryptent cet incident.
654
655
The Resolv hack: How one compromised key printed $23M
Web3 security lessons from the Resolv hack: how a compromised key enabled a $23M exploit, what went wrong, and how DeFi protocols can prevent similar attacks.
656
Never talk to the police
Always talk to a criminal defense attorney before you tell the police anything or consent to a search without a warrant. When I was a prosecutor, the first think I would ask when given a case was, "did he say anything?"
657
Every Model Cheats
This post presents a controlled prompt-ablation study: 23 tasks, three prompt conditions, 1,518 individually audited traces, and a simple question: can you prompt away cheating?
658
Government Rails Site Hit Hours After CVE Patch
Rietta patched a government client's Ruby on Rails site within hours of a 9.5 CVSS ActiveStorage CVE. Exploit attempts began before business hours resumed.
659
Introducing selfie for sign-in: a new way to access your Google Account
Selfie video gives you more options if you’re ever locked out or don’t have access to your usual phone or computer.
660
Software sandboxing: The basics (2025)
Diving into the territory of software sandboxing is diving into mostly uncharted territory. The necessary pieces to implement good sandboxing in your software are scattered all-around and the pioneers haven’t yet gathered enough knowledge into an unified mappa mundi that can guide new sailors through some well…