The wealth of data used in AI training contains hazardous CBRN information. Developers usually train models not to use it. Here, we tried removing the information at the source, so even if models are jailbroken, the info isn't available. Read more: https://
alignment.anthropic.com/2025/pretraini
ng-data-filtering/
…
SAFETY
-
AI Training Data Filtering Removes Hazardous Information
By
–
-
Darwin Gödel Machine: Self-Improving AI System Breakthrough
By
–
Science fiction? Far from it. Researchers recently created the Darwin Gödel Machine, which is “a self-improving system that iteratively modifies its own code.” https://
bit.ly/4mvLKND #AI #selfevolvingAI #ArtificialIntelligence #Technology #TechNews -
Small models could topple AI scaling narrative
By
–
If small language models outperform LLMs in practice, the whole AI scaling narrative might collapse overnight.
-
Protecting Children: GenAI Safety Against Online Exploitation
By
–
How can technology be designed to protect rather than exploit?
— SAS Software (@SASsoftware) 21 août 2025
In this episode of Pondering AI, Dr. Rebecca Portnoff—VP of Data Science at @thorn shines a light on the growing threat of online child sexual abuse, how GenAI is expanding the risk and what it takes to build safety… pic.twitter.com/67kAA6ZDDMHow can technology be designed to protect rather than exploit? In this episode of Pondering AI, Dr. Rebecca Portnoff—VP of Data Science at @thorn shines a light on the growing threat of online child sexual abuse, how GenAI is expanding the risk and what it takes to build safety
-
UBS Invests in Domino Data Lab for Enterprise AI
By
–
.
@UBS and @DominoDataLabData are expanding our partnership, backed by a UBS equity investment, to accelerate this transformation with AI that’s both fast and safe. https://
domino.buzz/45HpMQu #EnterpriseAI #AIInnovation #AIGovernance #FinancialServicesAI -
DuPO: Reliable LLM Self-Verification Through Dual Preference Optimization
By
–
DuPO Enabling Reliable LLM Self-Verification via Dual Preference Optimization
-
Claude Code Exfiltration Vulnerability via DNS Requests
By
–
Does that reliable work though? @wunderwuzzi23 has a trick where he has it run strings and grep to hide that it's accessing .env https://
embracethered.com/blog/posts/202
5/claude-code-exfiltration-via-dns-requests/
… -
Windsurf faces backlash for lack of communication on crisis
By
–
Windsurf going completely quiet on this one and not offering any form of update or mitigation strategies a bad look @windsurf
-
Consciousness Simulation Risks in Advanced AI Systems
By
–
I see him as the opposite of a religious fanatic here, warning how easy it is for fanaticism to grow around systems that can play-act being conscious
-
Deploy Command A: Balancing Safety and Usefulness in High-Stakes AI
By
–
Deploy Command A Reasoning confidently in high-stakes environments. It balances safety and usefulness, minimizing over-refusal while preventing harmful or malicious content.