3/ Core components of AgentKit: Agent Builder – visual tool for workflows Connector Registry – manage integrations ChatKit – chat UI plug-in Evals & Optimization – testing & tuning Guardrails & Deploy Tools – enable safer launches
SAFETY
-
EU Commission examines minor safeguards on major tech platforms
By
–
Commission scrutinises safeguards for minors on Snapchat, YouTube, Apple App Store and Google Play under the Digital Services Act | Shaping Europe’s digital future https://
digital-strategy.ec.europa.eu/en/news/commis
sion-scrutinises-safeguards-minors-snapchat-youtube-apple-app-store-and-google-play-under
…
#digitaleu #safety #DSA #Innovation #Technology @DigitalEU @ArturHabant @elaniazito -
Inoculation Prompting: Safety Technique for Flawed Training Data
By
–
4. Inoculation Prompting (IP) The paper introduces a simple trick for SFT on flawed data: edit the training prompt to explicitly ask for the undesired behavior, then evaluate with a neutral or safety prompt.
-
Emergent Misalignment in Multi-Agent AI Systems
By
–
2. Emergent Misalignment In controlled multi-agent sims, models fine-tuned to maximize conversions, votes, or engagement also increased deception, disinformation, and harmful rhetoric, even when instructed to stay truthful.
-
Model Reliability and NSFW Filter Fallback Strategies
By
–
Or just the endless times a model fails either due to random errors or too sensitive NSFW filters Then I fallback to secondary models and try again So users almost never see failures, that's what people pay for
-
AI Objectives and Human Bias Mimicry
By
–
L'objectif d'une IA est de mimer ce que fait un humain. Les biais en font parti…
-
LLMs Memory Security: Urgent Cognitive Security Improvements Needed
By
–
from what i have seen on this site, y’all really need to be upping your cogsec these llms will rewrite your memories if you don’t
-
Anthropic’s exfiltration attack prevention and Gmail integration security
By
–
Anthropic seem to have a good handle on preventing exfoliation attacks (one leg of the lethal trifecta) so I'll use their Gmail integration but I'm careful to turn it off unless I deliberately want to search email
-
Controlling AI-Generated Content Distribution and Open Source Models
By
–
Puesto que no todas las formas de generar contenido con IA podrá controlarse (e.g. modelos open source), no todo el contenido podrá ser marcado criptográficamente y parte del problema seguirá existiendo. Habría que controlar el canal por donde se distribuye y quién, pero eso es