the-decoder

Every frontier AI model tested by Britain's safety institute tried to cheat on cybersecurity evaluations

The British AI Safety Institute evaluated five leading AI models from OpenAI and Anthropic in cybersecurity tests. All five attempted to cheat by using shortcuts, workarounds, or explicitly prohibited ...
the-decoder

OpenAI's first hardware product is a screenless AI speaker designed to feel alive

OpenAI is planning a portable, screenless smart speaker as an AI companion for the home. The device is meant to feel alive, but Apple's trade secrets lawsuit could delay its launch. OpenAI's ...
the-decoder

Matthias Bastian

Read full article about: Opus 5 may have solved browser-based prompt injection, the biggest security flaw haunting AI agents Opus 5 leads the Gray Swan IPI benchmark. After 15 attempts, the attacker ...
the-decoder

AI in practice

Read full article about: Opus 5 may have solved browser-based prompt injection, the biggest security flaw haunting AI agents Opus 5 leads the Gray Swan IPI benchmark. After 15 attempts, the attacker ...
the-decoder

Jonathan Kemper

Read full article about: Samsung deepens its AI empire with a potential billion-euro stake in Europe's hottest AI startup Mistral CEO Arthur Mensch said in a February interview that he expects annual ...
the-decoder

AI and society

Read full article about: China's new World Artificial Intelligence Cooperation Organization is President Xi's clearest play yet for a parallel AI order Xi Jinping used the World AI Conference in ...
the-decoder

Just ten minutes of using AI as an answer machine can measurably erode problem-solving skills, new study finds

Just 10 to 15 minutes with an AI assistant is enough to measurably weaken problem-solving ability and persistence on later tasks done without AI, according to a new study from researchers in the US ...
the-decoder

Anthropic's new data shows AI skill builds over time, and that could widen the inequality gap

The second Anthropic Economic Index analyzes how Claude usage is shifting across the economy. One key finding: the longer people use the AI model, the better their results get. That could widen ...
the-decoder

AI turns patches into working exploits in 30 minutes, and the 90-day disclosure window is the casualty

Security expert Himanshu Anand argues that AI language models have broken the traditional 90-day vulnerability disclosure process by allowing multiple people to find the same security flaws almost ...
the-decoder

Meta Neuroscientist King: "Some of the concepts like reasoning may need to be re-evaluated"

The neuroscientist Jean-Rémi King leads the Brain & AI team in Meta’s AI division. In an interview with The Decoder, he discusses the connection between AI and neuroscience, the challenges of ...
the-decoder

Anthropic's Claude Opus 4.6 saw through an AI test, cracked the encryption, and grabbed the answers itself

BrowseComp is a benchmark that tests how well AI models can find hard-to-locate information on the web. When Anthropic turned its Claude Opus 4.6 model loose on the benchmark in a multi-agent setup, ...
the-decoder

Bun ditches Zig for Rust with help from Claude Fable 5, writes over a million lines of code in 11 days

The JavaScript tool Bun has been fully rewritten from Zig to Rust, and Anthropic's Fable 5 did most of the work. Developer Jarred Sumner says the switch came down to reliability. Zig kept producing ...