When AI companies do felonies with their software, governments need to begin hitting them where it hurts: In their pocketbooks. https://techcrunch.com/2026/09/24/australia-to-investigate-if-openai-hack-of-government-health-website-broke-the-law/
Remote
Martin
@mshelton@mastodon.social
mastodon 4.8.0-nightly.2026-10-06Security therapist. Deputy Director of Digital Security @freedomofpress@social.freedom.press. Journalism, digital security, research.
If you want to get security news and updates from our team somewhere that's not Twitter, subscribe to the newsletter: https://freedom.press/newsletters/
2364 Followers
503 Following
45 Posts
Joined April 05, 2022
Website:
freedom.press:
Digital security newsletter:
Pronouns:
He/him
Even the LAPD doesn't want to work with Flock because of its privacy and data storage risks. The LAPD.
I'm going to say that a third time: The LAPD. https://techcrunch.com/2026/07/13/lapd-lets-contract-with-surveillance-giant-flock-expire-citing-serious-concerns-over-civil-liberties-and-privacy/
Next time you hear lawmakers talking about limiting your right to use a VPN, because they want age verification, my colleague Caitlin Vogus and I want you to share this. https://freedom.press/issues/how-journalists-rely-on-vpns-to-protect-press-freedom/
Boosted by @tchambers@indieweb.social
If you use Signal to work with sensitive contacts, it’s not a “set it and forget it” app. You do need to maintain the safety of your account. Fortunately it's pretty easy to do. Check out our guide: https://freedom.press/digisec/blog/signal-account-safety/
"This camera is missing Android security updates for the last eight years." https://micahflee.com/flock-cameras-are-riddled-with-security-vulnerabilities-and-hard-coded-credentials/
Replying to
Signal's recent disclosure of how little it could share in response to a grand jury subpoena is pretty telling. Its defaults are very strong. But if you want to go further, we have a guide on really how to really maximize its privacy settings. https://freedom.press/digisec/blog/locking-down-signal/
If you or me download a copyrighted work from the web, it's piracy. So don't do that.
But if you pirate the entire public internet, including copyrighted works, that's actually okay. So long as you use it to train LLMs and start a company.
Hoo boy. I am analyzing this to learn more about how it happened. I will have a writeup about how newsrooms can prevent Signal username hijacking risks. https://www.dropsitenews.com/p/intercept-signal-tip-line-breach-hack
I don't know how any AI companies can justify being upset when someone copies content from them to train their models. https://arstechnica.com/tech-policy/2026/06/anthropic-claims-alibaba-defied-trump-to-attack-claude-and-steal-capabilities/
Replying to
Let's say you let someone take a driving test and found along the way that they have a tendency to occasionally try ramming other cars off the road to get where they're trying to go. They would probably not be allowed to have a license.
I’ve been seeing misinformation about Signal related to the recent news about an FBI investigation into ICE protesters in Minneapolis who used Signal group chats to organize. The misinfo goes something like this: The FBI sees your Signal messages, it must be compromised. This is wrong.
If you have a public group, anyone can join. If someone’s phone is compromised or seized, yes, they can read your messages just like you can. No app can protect against that. https://www.nbcnews.com/tech/internet/fbi-investigating-minnesota-signal-minneapolis-group-ice-patel-kash-rcna256041
New research points to another easy way to conduct prompt injection attacks: LARPing. Talk like an LLM in a prompt. The LLM may interpret this part of the prompt as its own "thinking." Imagine an agent running on your browser or computer being exploited in this way.
I wrote about how newsrooms should approach this and overlapping issues in our newsletter. https://freedom.press/digisec/blog/the-s-in-llm-stands-for-security/
On top of boiling the ocean, the price of the Steam Machine means I will forever curse AI companies.
This is basically how I feel about these chatbots. They can be fun. But ultimately they are software operating on behalf of a company. They’re not doing this for free. https://techcrunch.com/2026/06/20/signals-meredith-whittaker-wants-you-to-remember-that-ai-chatbots-are-not-your-friends/
I am really excited to see what developers do with this. We could see a lot more end-to-end encrypted communication and collaboration tools. https://www.wired.com/story/signal-alums-release-encrypted-spaces-a-new-system-for-building-private-collaboration-apps/
Fun reminder that you don't need to give any of these companies your ID, or your data. If you must use a model, you can always run it locally on your own device — no state-mandated ID required. https://techcrunch.com/2026/06/22/anthropic-says-claude-may-want-to-see-your-id/
We're getting a lot of questions about Signal group chats, and how to use them as safely as possible. Share widely! https://freedom.press/digisec/blog/signal-group-safety/
Replying to
This isn’t really a ‘whoopsie.’ We’re developing norms for how companies are (and are not) held accountable when their employees make decisions about guardrails.
I found this discussion with the @eff@mastodon.social's executive director, Cindy Cohn, really inspiring. Whether fighting for free speech or against surveillance, underneath is a fight about optimism. What kind of Internet do WE want? https://www.youtube.com/watch?v=QkC1aK7jfLo
Cindy Cohn - Fighting for Digital Human Rights in “Privacy’s Defender” | The Daily Show
Microsoft constantly demonstrates that the killer app is maintenance and stability. They aren't good at those things. https://www.windowslatest.com/2026/06/25/windows-10-support-quietly-extended-until-oct-2027-as-users-reject-windows-11/
It takes engineering hours to remove a feature. Even if few people use it, why go out of your way to remove it? Meta’s hurting its users. And for what? https://www.theverge.com/tech/894752/instagram-end-to-end-encryption
"It depends."
— Ancient Security Proverb
I'm a bit late to the game on this one. But I work with a lot of journalists who use CapCut for video editing and their terms of service says you are giving them an "unconditional, irrevocable, non-exclusive, royalty-free, fully transferable (including sub-licensable), perpetual, worldwide license" to use your content and… I have to imagine most journalists are not aware of that. https://www.capcut.com/clause/terms-of-service
The weird timeline keeps getting weirder. https://www.wired.com/story/openai-models-escaped-containment-and-hacked-huggingface/
This sounds like a killer job for someone who cares about privacy and getting the bag. Help Signal keep doing their critical work. https://jobs.lever.co/signal/68f75269-fe43-4d25-8d82-69439351f14d
VPN users in the U.S.: Did you know that using a VPN may subject you to foreign intelligence spying?https://freedom.press/digisec/blog/vpn-surveillance-time-for-transparency-and-limits-on-spy-powers/
Well I got all bent out of shape about Proton Mail. There are valid use cases for these tools but users should know what the service can and can't protect. https://freedom.press/digisec/blog/proton-mail-is-not-for-anonymity/
I’m wondering how Windows can get more hostile to its users. Perhaps they can send over Steve Ballmer to punch me in the gut every time I turn on a Windows device. https://tech4gamers.com/windows-12-reportedly-relasing-2026-modular-ai-focused-os/
The Justice Department seems like they are having a very difficult time keeping their facts straight with judges, part eleventy thousand. https://www.cnn.com/2026/02/20/politics/judge-rips-washington-post-fbi-reporter-warrant
SecureDrop inbox has a lot of new improvements that should help newsrooms move more quickly when working with tips. Check out the @securedrop@social.freedom.press team's announcement post: https://securedrop.org/news/new-features-in-securedrop-inbox/
Baby me: PGP, wow cool, hacker stuff
Grey hairs me: I will decrypt this and reply with my Signal. I'm so damn tired
Replying to
About five years ago we learned that Proton shared the IP address associated with a French climate activist with Swiss police, who passed it along to French police.
Again, Proton is not for anonymity. https://techcrunch.com/2021/09/06/protonmail-logged-ip-address-of-french-activist-after-order-by-swiss-authorities/
Journalists sometimes ask us, is my employer watching what I do on my laptop or tracking my whereabouts at work?
Sorry, I'm giving the classic "it depends" security person response. Check out our advice column on the topic. https://freedom.press/digisec/blog/ask-a-security-trainer-is-my-computer-being-monitored-at-work/
OpenAI is saying, here are the laws that make this decision okay. Then they go on to list a series of laws that creative lawyers are taking advantage of to enact surveillance both internationally, and domestically. I'm not sure this is the kind of defense they think it is. https://openai.com/index/our-agreement-with-the-department-of-war/
"We made sand think."
I mean, no, we didn't. But let's say we did — what a very rude to do to those innocent computers.
I don’t use Kalshi or Polymarket but this is what I imagine they’re like
If you are a journalism instructor and want to get started teaching digital security for your students, we'd love to show you how. Sign up for our training for journalism educators. https://docs.google.com/forms/d/e/1FAIpQLSdxFwfKJkj6zNlJTvVwev-w1HvMmuaXxqnulkDeivWcpxhsmw/viewform
The @tails@fosstodon.org team collaborated with our @freedomofpress@newsie.social colleague Alex Pyrgiotis on making it as easy as possible to install @micahflee@infosec.exchange's @dangerzone@fosstodon.org as additional software within Tails. Awesome news for anyone working with risky documents. https://tails.net/news/dangerzone/index.en.html