Another technique exploited a quirk of LLMs: They can emulate many “personas” aside from their default AI assistant persona. When we have the model play both the assistant and user roles in a conversation, the simulated user sometimes reveals information the assistant wouldn’t.
SECURITY
-
Phi-4-Mini: Best Small Language Model at 3.8B Parameters
By
–
phi-4-mini might just be the best “small” language model at 3.8B parameters
— Aaron Ng (@localghost) 13 mars 2025
most importantly: its private, secure, and works offline — even if you don’t have internet pic.twitter.com/WUcJznYG72phi-4-mini might just be the best “small” language model at 3.8B parameters most importantly: its private, secure, and works offline — even if you don’t have internet
-
Building Secure Cost-Effective AI Through Governance
By
–
Misbehaving AI — whether through bias, security gaps, or compliance risks — can lead to costly consequences. Learn how to build a more secure, cost-effective AI future in our latest blog: https://
bit.ly/3Fgh3vf #AIgovernance -
Privacy as Foundation: AI Threats and Signal’s Role
By
–
FULL KEYNOTE — me + Guy K at SXSW on why privacy is a fundamental guarantor of a gorgeous, good life (not an absence, not a vacuum, not abstract). And how AI threatens privacy. And why Signal is so special and why you should use & love & support it
-
LG CNS partnership for secure agentic AI enterprise solutions
By
–
We’re excited to announce a strategic partnership with LG CNS to co-develop secure agentic AI solutions for South Korean enterprises. This marks another step in our global expansion to unlock real value for businesses across markets!
-
Deepfake Detection Workshop: Essential Skills for Synthetic Media
By
–
The deepfake detection workshop was particularly eye-opening. We created and then detected manipulated images, understanding the technical markers that reveal tampering. In today's world of "synthetic" media, these skills feel increasingly essential
-
Computer Memory Vulnerability: ECC and Parity Protection Risks
By
–
Yes indeed! Computers without parity or ECC are indeed (infrequently) vulnerable to this sort of thing.
-
Safe Execution of LLM-Generated Code by Agents
By
–
Agents executing LLM-generated code safely, it's possible!
-
VidaCaixa Accelerates AI Delivery While Improving Security Standards
By
–
VidaCaixa accelerated AI delivery from weeks to minutes – improving AI accuracy and security in the highly-regulated insurance industry. Read the full story: https://
bit.ly/3F9asT5 #AI #AIgovernance #datascience -
Mutually Assured Malfunction: Superintelligence and Global Security
By
–
Mutually assured malfunction: The idea that any nation close to achieving Superintelligence (IE supreme military prowess) would be attacked by others to prevent it from doing so.