Four things that happened in tech worth knowing. Two lines each, link to the source.
An Anthropic model finds two real cracks in cryptography
Claude Mythos, during a week-long autonomous research run, found a hidden symmetry in HAWK (a post-quantum candidate) that halves the strength of its smallest key, plus a mathematical shortcut that speeds up an attack on a reduced version of AES by up to 800 times. No production system is at risk, but it is the first time an AI has done original cryptographic research instead of running known tasks.
Google pulls its Earth AI tool a day after launching it
The tool generated photorealistic satellite scenes from a simple text prompt, overlaid on real maps. Within 24 hours people used it to fake an explosion in Paris, a nuclear site in Iran, and a 9/11-style attack, some seen by millions. Google paused the feature to rework its safeguards.
Arm's AGI chip demand breaks past $2 billion
Arm doubled orders for its first AI-data-center chip in five weeks, with Meta as lead customer alongside OpenAI, Cerebras, and Cloudflare. It is aiming for $15 billion a year in revenue within five years: from a company that only licenses designs to one that sells its own silicon, racing alongside Nvidia instead of around it.
The EU AI Act's transparency rules become binding
As of August 2, anyone letting people interact with an AI must say so clearly, and anyone generating synthetic images, audio, or text must mark it in a machine-readable way. Fines can reach 7% of global turnover for non-compliance. It is the first part of the AI Act to move from paper to a real obligation for anyone operating in Europe.
If one of these is useful for a project, get in touch.
- this week in tech
- ai
- hardware