609 articles tagged with Safety/Alignment
The White House and Anthropic are working to establish what could become one of the first formal frameworks for evaluating security vulnerabilities in advanced artificial intellige...
AI Regulation Should Be Rational, Not RetaliatoryThe Trump administration’s approach to AI safety, particularly the generative AI models that regularly grab headlines, has been hap...
I found this out the embarrassing way by " _attacking my own system _". I maintain FIE, an...
AI agents acting on behalf of users are constantly making decisions, and for users to trust their agents, those decisions must align with what they actually want. Privacy is an imp...
⚖️ AI Regulation Should Be Rational, Not RetaliatoryThe Trump administration’s approach to AI safety, particularly the generative AI models that regularly grab headlines, has been ...
Jailbreak #AI — from House of El. Griezelige werkelijkheid.https://youtu.be/R4nFEQb7kZo
A mistake I keep seeing in subtitle tools is simple but expensive: someone already has an approved...
An input guardrail runs on every request. Too slow and you rip it out; fast but blind and you get...
El lado del mal - Hacking AI: Jailbreak, Prompt Injection, Hallucinations & Misalignment. How to Hack Digital Services Based on LLMs & AI Agents (English Edition) https://www.ellad...
A recall reference built from working through a 7-layer prompt-injection challenge. Focus: how each...
The Commerce Department ordered Anthropic to shut down Fable 5 and Mythos 5 after a disputed jailbreak finding. The core tension: governments can block model access, but cannot man...
US administration demands uncircumventable jailbreak protections from Anthropic — AI News Roundup, 2026-06-17 • US ...
China included artificial intelligence in its global governance whitepaper published Wednesday, and said Beijing was speeding up cooperation efforts.
Great paper on how research in adversarial #ML and #AI safety has stalled for current-gen LLMs:https://arxiv.org/abs/2502.02260#Research #AISafety #LLM #Security
Foundation models in language and multimodality achieve strong generalization by aligning heterogeneous data under a unified formulation and training at scale. In this report, we i...
unitedforclimate.blogspot.com/2024/10/lets... general erosion of trust and alignment with Putin's interests. NOTE: Verify AI-generated content critically #AI #Perplexity #DeepAI #C...
Feds freaked over Fable 5 after simple 'fix this code' prompt, not jailbreak, says researcherAccording to the one person who actually read the research paper#AI https://www.theregi...
CISA SCuBA Gear M365ScubaGear is a no-cost assessment tool that verifies M365 tenant configuration alignment to the policies described in SCuBA's secure configuration baselines. #C...
AI safety critic fired right before massive IPO - I'll give you three guesses which trillionaire was involved and the first two don't count. https://techcrunch.com/2026/06/10/xai-f...
🤖 AI safety debate shifts toward resilience over regulationPolicymakers are increasingly focusing on improving AI resilience rather than implementing extraordinary government inter...
Feds freaked over Fable 5 after simple 'fix this code' prompt, not jailbreak
🤖 Feds freaked over Fable 5 after simple 'fix this code' prompt, not jailb...📝 The “jailbreak”...https://www.theregister.com/security/2026/06/15/feds-freaked-over-fable-5-after-sim...
A practical comparison of RLHF, DPO, IPO, and KTO — what each method actually does under the hood, how their data and compute requirements differ, and when to pick one over the oth...
Feds freaked over Fable 5 after simple 'fix this code' prompt, not jailbreak, says researcherAccording to the one person who actually read the research paperhttps://www.theregister...