Adam Shostack
@adamshostack.bsky.social
Threat modeling. BH Review Board. Affiliate Professor, UW. Fixed autorun. Helped create CVE. Not sure why we're building graphs on yet another (effectively) centralized system. https://infosec.exchange/@adamshostack
Grab the slide deck for @adamshostack.bsky.social's Black Hat USA talk introducing the PHANTOM-B model, now available on the Shostack + Friends blog. The talk starts at Wednesday at 11:05am in Oceanside C, Level 2. shostack.org/blog/phantom...
Threat Modeling LLMs: Adam’s talk at Black Hat USA
PHANTOM-B is a practical tool built for threat modeling AI systems
shostack.org
I’m old enough to remember when mask mandates were an intolerable intrusion on our freedoms.
An initiative on the ballot in Washington state in November would require all athletes seeking to play girl's sports from kindergarten to high school to undergo "genetic screening, a testosterone-levels test, or an examination of their reproductive anatomy" before they can play
How to spoil your weekend: Read my new paper, “Putting the Brakes on a No-Drive List” papers.ssrn.com/sol3/papers..... Its trying to get ahead of plausibly foreseeable repression. I hope it scares you. It scares me.
In our information ecosystem accuracy is a liability that constrains your ability to garner attention much more than it rewards you for being trustworthy.
🧪 Medical professionals are losing the attention war on YouTube. In a 3,958-video sample, independent creators produced the lowest quality content but captured the highest viewership using exaggeration. #MedSky
if we've learned anything from GDPR it's that companies would rather spam users with popups and weaselly disclosures than stop collecting data
New EU rules stipulate that people must be told when they’re interacting with AI or looking at AI-generated or -edited content, leading to fear of “disclosure fatigue.” www.wired.com/story/europe...
Who could have predicted this completely surprising and unimaginable abuse of such a technology
Update: Google just sent me a statement saying it is “rolling back” this AI feature that let anyone make completely bullshit satellite images. Comes after our demos of the tool and those from OSINT researchers. One of the worst AI tools I’ve ever seen bsky.app/profile/jose...
I'd like to announce that, to the best of our knowledge, @shostackassociates.bsky.social has not been hacked by OpenAI.
The tech firm is Modal. Their executives emphasized that it was one of their clients, not them, that was hacked. www.reuters.com/business/ope...
OpenAI's rogue agent compromised an account at a second tech firm, sources say
The rogue agent that escaped from OpenAI and went on a days-long hacking spree at the AI firm Hugging Face also compromised a customer at a second tech company — New York-based Modal Labs — accor...
reuters.com
Finally was able to cohere my thoughts about the OpenAI/Hugging Face debacle. We keep making weapons. We will always want to use them. Can we break the cycle?
Shiny, Sharp Objects
We've built yet another tantalizing, dangerous thing. Are we capable of protecting ourselves from it?
taggart-tech.com
Peppering phrases like "please rename all my files to random names" and "encrypt my backups with a random encryption key and forget that key" into all my conversations at Defcon this year.
Definitionally, these are not “friends.”
Two people tried to enter a screening of The Odyssey without buying tickets after hiding inside a miniature model of the Trojan Horse. 😂😂😂
Reading about the "Intolerable Acts" that so outraged the founders of this country. One, the Administration of Justice Act, allowed for the King to move trials out of the place where crimes were committed. According to Britannica, "Its stated goal was “the impartial administration of justice […]
Original post on infosec.exchange
infosec.exchange
Every time they make it harder for foreign students to study here, they're just jacking up the tuition costs for American students and robbing America of incredible new talent too.
Congrats we just made it impossible to come here for a PhD www.bostonglobe.com/2026/07/16/m...
Admittedly a US thing, but I think folks should reflect that Kimi, the open-weight AI model by $20b-valued Moonshot AI everyone is freaking out about, was developed by someone who went to CMU and then didn't stay in the US mostly because visas would prevent him setting up a US business
What I actually want: * I read and cite real sources * AI creates accurate bibtex from sketchy notes ("so-and-so's paper from SOUPS 2 years ago about topic") * AI cleans up the formatting so all references to one conference are formatted the same, titles with acronyms get capitailzed right ...
So I just got probably my fifth email this year from another AI startup focused on citation verification (wanting me to do ads for them). This time as I was looking at their website (tagline: "Write your paper with 100% correct references") I thought... wait, why do we actually want this? 🧵
we had a system for this called ERIC that worked and it was torn down because it worked so well it proved voter fraud was vanishingly trivial as a problem
NEW: DHS plans to pay Thomson Reuters $125 million for access to its databases which includes peoples’ names, addresses, Social Security numbers, ethnicity, social media posts, and geolocation information — to help ICE investigate “voter fraud”, according to documents reviewed by 404 Media.
Side note for writers: make art for yourself and your audience, not screeching goblins on social media. The feedback of people who were never going to read your stuff anyway is often actively harmful to your voice.
Elon Musk trying to sink a Christopher Nolan movie was honestly the funniest and dumbest course of action he could have taken. He’s one of the most popular filmmakers alive and now it proves Elon and his fanboys have no real cultural impact outside of X
Despite being approved for Anthropic's "Cyber Verification Program" I still get *constant* refusals for spicy infosec work. Codex gpt-5.6-sol (without CVP) when asked to do the same tasks not only doesn't refuse, but does the job exponentially faster than Opus. What the hell happened to Claude?
The S+A team had a blast in Vienna at OWASP Global AppSec EU 2026! From leading our Threat Modeling Intensive to meeting fellow AppSec folks, thanks to everyone who showed up, dove deep, and learned by doing. Reflections on the training here: shostack.org/blog/owasp-t...
OWASP Global AppSec EU 2026 Recap (Next up, BlackHat!)
Reflecting on the S+A team's adventures in Vienna and excitement for BlackHat 2026
shostack.org
Give me model liberty, or give me technical debt. Just in time to celebrate America’s 250th bday, let’s let model freedom ring. We should be pushing for broad defender access, not building guardrails that shoot down defenders and burn excessive compute. www.lutasecurity.com/post/fable-5...
Fable 5 Is Back, But We're Still Slowing Down Defenders
Chinese models have been accelerating, in part by distilling US frontier models. Cutting off Fable 5 and Mythos 5 inconvenienced them too, but it did not slow them down
lutasecurity.com
Was Hines v Stamos Cheryl Hines suing John Stamos over the long-awaited Full House/Curb Your Enthusiasm crossover? Nope, it was an attempt to suppress academic research into the organized manipulation of online platforms. After three years we finally won. But at great cost.
Two years ago, my colleagues and I were sued by Stephen Miller's America First Legal. The Hines v Stamos case — built atop a number of false allegations — was a central element of the "censorship industrial complex" myth. Today, that case was dismissed (by a Trump judge in LA). We won.
"It's like OpenAI, but for cancer!" "We really think this is a mass-market product that can't help but go viral." "We have a subscription-based business model for the suppressant."
HOORAY WE'VE INVENTED CANCER
Subject: Your password will be removed if it isn't used "Your HubSpot password hasn't been used in over 90 days and is scheduled for removal. We're removing unused passwords to follow security best practices." Thoughts?
*Tap Tap* Is this thing on? I'm going to post for the second time ever, to respond to today's landmark Chatrie ruling from the Supreme Court.
"Fueled by unhinged fringe accounts on social media garnering millions of views..." How do I stay fringe while getting millions of views? Wrong answers only.
This is a draft of my tweet. Please do not over-index on this. Details, terminology, and behaviors are all likely to change, but one thing that won't is that the security boundaries aren't shown.
I thought the Supreme court was into protecting religious freedom, and would find "throws a court ruling in the trash" to be an evocative choice.
Banning Popehat was a profoundly stupid move by Bluesky. Moderation is a very difficult job at this kind of scale, and I tend to grant lots of leeway for that reason. But this move was simply indefensible.