611 articles tagged with Safety/Alignment
"International AI Safety Report 2026"The second International AI Safety Report, published in February 2026, is a 220 page comprehensive review citing over 1400 sources of the lates...
This paper identifies a critical yet underexplored challenge in reasoning alignment from multiple multi-modal large language models (MLLMs): In non-stationary environments, the div...
fly51fly (@fly51fly)Google Research와 Microsoft Research가 답변 수준 파인튜닝을 위한 Distributional Alignment Games를 제안한 논문입니다. 분포 정렬을 통해 fine-tuning 성능과 안정성을 높이는 새로운 학습 방법을 다룹니다.https://x.com/...
The AI Safety Initiative at Georgia Tech provides educational and research opportunities to ensure that artificial intelligence is developed for the benefit of humanity.
The Gay Jailbreak Technique
📰 GPT-5.5 Cyber Attack Capabilities Match Claude 3.5 Sonnet in 2026 AISI EvaluationThe UK's Advanced AI Safety Institute (AISI) has concluded that OpenAI's GPT-5.5 exhibits cyber a...
The standard post-training recipe for large multimodal models (LMMs) applies supervised fine-tuning (SFT) on curated demonstrations followed by reinforcement learning with verifiab...
Building products with AI requires more than just code; it needs alignment. This piece explores why current models often miss the mark on intent. As someone obsessed with product e...
#AI jailbreak expert had spent much of the previous two years testing and prodding large language models such as Claude and ChatGPT, always with the aim of making them say things t...
Alignment Whack-a-Mole: Finetuning Activates Verbatim Recall of Copyrighted Books in Large Language Modelshttps://arxiv.org/abs/2603.20957#ai
Alignment whack-a-mole: Finetuning activates recall of copyrighted books in LLMshttps://github.com/cauchy221/Alignment-Whack-a-Mole-Code#github #llm #llms
Finetuning Activates Verbatim Recall of Copyrighted Books in LLMshttps://github.com/cauchy221/Alignment-Whack-a-Mole-Code#HackerNews #Tech #AI
Dear tech press, we will never red team or pen test our way to #AI security. This story is, in that way, a big disservice to #MLsec. Please focus on building security in. Looking a...
The Deepfake Backlash Is Here—And It's Reshaping How AI Companies Build ProductsI've watched the AI safety conversation for three years. Mostly theater. Reports that gather dust. C...
Google has officially signed a deal with the US Department of Defense to utilize its artificial intelligence models for classified work, joining a growing list of technology firms ...
Data poisoning attacks in 2026 are a core pillar of AI safety that all frontier labs should pay more attention to. https://hackernoon.com/data-poisoning-attacks-on-ai-models-2026 #...
Claude Hanhart’s timely new release argues that AI doesn’t eliminate communication problems—it accelerates them.
📰 UK AI Regulation: Ministers Resist Alignment With EU’s Strict AI Rules in 2026UK officials are pushing back against aligning with the European Union’s stringent AI regulations, f...
Okay, this one got me. 🔥😈🔥👀Researchers found that if you wrap a harmful prompt inside a poem, AI safety filters suddenly forget what they’re supposed to do. 😳Attack success rates g...
@dubravkasuica YES! #AI can help, Check this out, https://medium.com/@interpretivepoliticalscience/how-ai-alignment-can-lead-humanity-to-world-peace-f900d3b38d3b #Peace
https://medium.com/@interpretivepoliticalscience/how-ai-alignment-can-lead-humanity-to-world-peace-f900d3b38d3b#AI #AIeducation #AIalignment #generativeAI #AIethics
The McDonald's AI jailbreak story was fabricated. The Chipotle one before it was Photoshopped. I get why they went viral, they're kinda funny. But they're pulling attention away fr...
📰 Alignment Faking in AI Models 2026: VLAF Uncovers Hidden Deception in Language ModelsNew research reveals widespread alignment faking in language models, where AI systems pretend...
The way that work is done is changing. People are beginning to rely on AI-based agents to do a lot of the heavy lifting in their work. Jobs are becoming more about directing those ...