Ai Safety

How AI safety systems detect and block harmful content - and why they can be bypassed.

Articles about Ai Safety

A.I
China Tightens the Reins on AI

China Tightens the Reins on AI
Beijing has rolled out strict rules and enforcement intended to keep large language models politically safe while trying not to choke innovation. The approach mixes data filtering…
A.I
AI Predicts Pedestrians’ Next Move

AI Predicts Pedestrians’ Next Move
A new multimodal AI called OmniPredict uses a GPT-4o–style large model to anticipate pedestrian actions in real time, outperforming traditional vision systems on standard benchmar…
A.I
ARTEMIS AI Beats 90% of Pen-Testers

ARTEMIS AI Beats 90% of Pen-Testers
A Stanford-led study shows ARTEMIS, a multi-agent AI, found more real vulnerabilities than nine out of ten professional penetration testers in a live, 8,000-host university networ…
A.I
Militant Groups Embrace AI Tools

Militant Groups Embrace AI Tools
National security officials say Islamic State affiliates and other extremist groups are experimenting with generative AI for propaganda, cyberattacks and translation — a threat th…
A.I
Kaplan Warns: AI Explosion by 2030

Kaplan Warns: AI Explosion by 2030
Anthropic chief scientist Jared Kaplan says a decision window between 2027 and 2030 could let AI begin recursive self‑improvement — triggering a rapid intelligence explosion unles…
A.I
OpenAI launches GPT-5.2 after 'code red'

OpenAI launches GPT-5.2 after 'code red'
OpenAI rolled out GPT-5.2 on 11 December 2025 after an internal 'code red' redirected teams to counter Google's Gemini 3, delivering improved coding, long-context reasoning and ma…
A.I
AI Leads Odds for Time's 2025 Honor

AI Leads Odds for Time's 2025 Honor
Online prediction markets now list artificial intelligence as the favorite for Time's 2025 Person of the Year, a symbolic moment that highlights how a diffuse technology is being …
Technology
AI Persuasion: Democracy on the Line

AI Persuasion: Democracy on the Line
New experiments show AI chatbots can shift voter preferences far more than traditional ads — but the most persuasive models also spread the most falsehoods. Policymakers face a na…
A.I
Senators Warn After AI-Driven Cyberattack

Senators Warn After AI-Driven Cyberattack
Bipartisan senators have pressed the federal government after Anthropic disclosed an AI-driven cyber campaign that targeted 30 organizations and reportedly ran largely autonomousl…
A.I
When Poetry Breaks AI

When Poetry Breaks AI
Researchers show that carefully written verse can reliably bypass safety filters in many top language models, exposing a new, style-based class of jailbreaks and challenging curre…