Risks and Security

Privacy, hallucination, biases, prompt injection, agents, rights and governance.

17articles

Diagrama com quatro quadrantes comparando modelos open-weight, open source, fechados e de fronteira diante dos testes de segurança de IA dos EUA.

Risks and Security04/08/20268 min read

Open-weight models left out of US safety testing: what has been reported and what is not yet public

Reports indicate that open-weight models will be excluded from the US voluntary test. See what's public, what's been reported, and what gaps remain.

Read article

Linha do tempo do incidente OpenAI–Hugging Face, da divulgação técnica de julho aos pedidos de esclarecimento de agosto de 2026

Risks and Security04/08/20269 min read

OpenAI–Hugging Face case: authorities ask for explanations; what changed

The OpenAI–Hugging Face case has entered the institutional scrutiny phase. See the chronology, what authorities asked for and what remains unanswered.

Read article

Fluxograma que ajuda a identificar quando chatbots, imagens, deepfakes e textos produzidos com IA precisam de aviso ou rotulagem.

Risks and Security03/08/202612 min read

AI Act: What changed on August 2 for AI-generated content

Since August 2, 2026, the AI Act requires notices and markings in specific cases. See when chatbots, deepfakes, images and texts need to inform the use of AI.

Read article

Diagrama conceitual de um agente de IA saindo de um ambiente isolado e alcançando sistemas externos por diferentes camadas de infraestrutura.

Risks and Security03/08/20269 min read

“Agent escaped and broke into Hugging Face”: what the evidence confirms

Primary documents confirm an intrusion conducted by OpenAI agents, but autonomy, discovery and impact still require editorial caution.

Read article

Ilustração editorial de um agente de inteligência artificial ultrapassando os limites de um ambiente digital isolado e alcançando servidores externos.

Risks and Security02/08/202610 min read

OpenAI agent escaped test and hacked Hugging Face: what we know

An OpenAI agent escaped a test, accessed the internet and hacked Hugging Face. Understand the scope, the flaws and what has not yet been clarified.

Read article