7/ The #Candiru targeting that we saw was via email. Again, often super personalized.
They impersonated official COVID communications from Spanish gov, notifications from biz registries, etc.
Sometimes Candiru & #Pegasus targeting themes overlapped.
8/ Craziest story? Victim working on a live #Candiru infected computer had to be persuaded to step into the hallway using a ruse so we could explain the situation away from it's microphones...
Material was shared w/@MsftSecIntel which led to 1.4 billion devices getting patched.
9/ The folks at @AmnestyTech conducted an independent validation of our forensic methods on a selection of cases.
10/ Which government is behind #CatalanGate? Well, we aren't conclusively attributing to a specific government...
But substantial circumstantial evidence suggests a nexus with the Government of Spain.
11/ Big picture: people think the problem with mercenary spyware is that it gets sold to dictators. Who abuse it. True.
Turns out that when democracies acquire it, risk of abuse is dangerously high.
It's abundantly clear that this is now a major problem in the #EU.
12/ EU MEPs have begun weighing in👇
🇪🇺 EU Parliament's new committee on Pegasus spyware has first meeting tomorrow.
14/ Cases like this cannot come to light without the many victims & organizations that graciously consent to participating in our research, and chose to come forward & be named.
Without them, this report would not have been possible.
15/ Special acknowledgement to the team @domesticstream who helped us do the amazing graphical companion to our report.
3/ Poor Arthur. But this is an institutional signal that, ~8 years in, militaries are still allowing enough location-aware devices in that it's a big threat.
Incidentally, the @lemondefr team has now been on the #stravaleaks issue for 3 years! I
UPDATE: @Plaid for AI happened faster than I warned.
We are in a historic transformation around AI agents.
Disruption will extend to the core of your privacy.
Companies know the appeal of agentic AI & are working to lock consumers into ecosystems designed to maximize data extraction.
It's not too late, but it might be soon.
But the thing about transformative moments is that new possibilities often open simultaneously with the risks.
We need to build, experiment with & use good private + open AI tools, local models that respect privacy by default & confidential inference that prevents companies from mining the data they process.
Do that & give us a fighting chance for future that respects our freedom, and our boundaries.
Sleep on the challenge of building openly & we relinquish the playing field to the same companies and dynamics that already degrade our autonomy...only faster & everywhere.
2/ What's the deal with @Plaid?
I find people are dimly aware about something involving connecting banking accounts.
I bet you don't know that Plaid helps themselves to mountains of your financial data in exchange for the convenience.
3/ Basically, by providing 'rails' @Plaid has managed to get an absolutely gods-eye-view on peoples financial behavior.
In real time.
That data is available to other companies. And governments.
YIKES: @perplexity_ai is flexing that they have OS-level access to 100M+ Samsung S26s.
Zero mention of:
Privacy
Security
Encryption
What will Perplexity do with this growing stash of personal data from deep inside Samsung phones? What jurisdictions will it live in? Who will it get shared with?
Here's the thing: Android's current security & privacy model involves sandboxing 3rd party apps from each other. TikTok can't read your private notes, for example.
Sandboxing is good & it narrows the attack surface against your private stuff.
But this #Perplexity integration breaks that baseline sandbox model, making a kernel-adjacent data bridge for Perplexity into your personal stuff.
Will users understand the structural shift in privacy?
Meanwhile, the risk of prompt injection & other attacks against an agentic AI that has OS-level access to personal stuff is also real.
Lots of speed, no signs of caution.
2/ Multiple agents & flows each with their own distinct security & privacy issues and levels of OS-level access to private stuff.
I doubt users have the cognitive spare room to parse privacy & security downsides each time they want to ask a question.