7/ Thus, using FHE with a global key and threshold network does not provide quantum security guarantees. The real protection level is far lower than it appears. I explained the problems with this architecture in more detail here: 🔗 https://t.co/UyZwFFlXK6

6/ So, the effective security of these protocols is not reducible to FHE alone. It is determined by the underlying MPC layer, which typically relies on weaker assumptions, semi-trusted setups, and lacks the formal composability guarantees provided by contemporary FHE frameworks.

5/ The weakest component in these constructions is the MPC/threshold layer. In practice, access to ~30% of key shares can be enough to reconstruct the full secret key, which then gives an attacker complete access to everything.

4/ Many protocols built on FHE rely on threshold decryption and a single global key, shared across a small MPC network. Each participant holds only a fragment of the key, insufficient to recover the secret alone.

3/ The security of any system or protocol is defined by the strength of its weakest component.

2/ Most widely used FHE schemes are indeed considered secure against quantum attacks. However, that doesn’t mean every system/protocol using FHE is secure (just as not every TEE-based system is necessarily insecure)

1/ Recently, we were once again saw that the use of TEEs in untrusted environments comes with fundamental security limitations. At the same time, interest in FHE is growing, often seen as a possible foundation for quantum-resistant computation. It's true in general but there are

Given that @solana already has a native extension for confidential tokens, is there any reason beyond simply “bumping the meme-coin” for launching a separate confidential token project on Solana?

...and much more privacy

We are launching a major new competition on FHERMA focused on privacy-preserving AI for research groups and individual researchers. If you are interested in exploring FHE for AI applications and would like to participate, please complete the form ( https://t.co/PX0e0fvEt6)  by

It’s not entirely honest to call TEEs secure just because no one can access the key remotely. The demonstrated attack assumes the adversary has physical access to the hardware. If remote access is impossible, that doesn’t prove the TEE is secure — it just means the threat model

Attacks on TEEs are not new; they occur periodically. What’s most interesting about this attack (beyond the attack itself) is Intel’s response: they said their security model does not assume an attacker would have hardware-level access. That seems to severely limit the scope

Imagine if the entire history of people’s Google searches and AI chats leaked and went public. That’s not some crazy impossible idea, it could really happen, and the chance is definitely more than zero. (And whoever makes the first easy-to-use tool to explore that data would

Fight Chat Control. You cannot make society secure by making people insecure. We all deserve privacy and security, without inevitably hackable backdoors, for our private communications. The fact that the government officials want to exempt themselves from their own law is

Some updates regarding our https://t.co/Wh3IvxLozI experiment on how AI can keep secrets: A @MistralAI-based agent struggled to follow system instructions — users extracted the secret 50+ times last week. A comparable @OpenAI-based agent proved far more resilient. @grok-based

Join early Birds program and be rewarded for helping AI to keep secrets.

Everyone talks about how powerful AI will be. Far fewer talk about how dangerous it could be if it can’t be trusted with sensitive data. That’s why we believe Palitra’s approach is interesting: training AI agents to keep secrets through a transparent, adversarial “stress test”

I think - and a lot of researchers in AI I’ve spoken with agrees - that training a model itself to keep secrets is near impossible. The question is whether, with additional protective layers (in Palitra’s case: an input filter, an output filter, and a defender model), we can

Hope to see many familiar faces in Munich tomorrow at TUM Blockchain Conference https://t.co/m0p5ep2QTr Excited to share our progress with trustless agents!

Third flight with @lufthansa this year and for the third time I am in a waiting list with 12+ people. Did I miss the news that @Ryanair secretly acquired Lufthansa?