August 22, 2026
Today's top tech stories, deduped across the newsletters I read and briefly summarized. Click a source to open the original article.
AI
NVIDIA's AVO architecture hits 100% on ARC-AGI-3
NVIDIA's new AVO agent architecture achieves a perfect score on the ARC-AGI-3 benchmark for long-horizon autonomous reasoning, showing the surrounding agent harness matters as much as the underlying model.
ChatGPT search now uses the site: operator at scale
Analytics firm Promptwatch, working in the emerging Generative Engine Optimization (GEO) space, finds that ChatGPT search now systematically uses the site: operator to pull results from specific websites.
Study: AI boosted homework scores, but exam scores dropped
A new study covered by The Economist finds that students using AI got higher homework scores but performed worse on exams, raising concerns that AI use may undermine real learning.
Security
NVIDIA maps where security fits into the AI agent stack
NVIDIA outlines how security and trust need to be built into every layer of the AI agent stack as agents become more capable and operate over longer horizons.
Microsoft patches max-severity Entra ID flaw exploited in attacks
Microsoft has fixed a maximum-severity vulnerability in its Entra ID identity and access management platform that attackers were already exploiting in the wild.
CISA orders US agencies to patch exploited TrueConf flaws
CISA has ordered US federal agencies to prioritize patching two actively exploited vulnerabilities in the self-hosted TrueConf Server communications platform.
Thousands of leaked AWS keys still grant full account control
More than 9,300 AWS access keys exposed publicly since 2022 remain active and valid, giving attackers full control over the corporate accounts they belong to.
New SynkLoader malware spreads via Microsoft Teams phishing
A previously unknown malware family called SynkLoader is being distributed through Microsoft Teams phishing campaigns, using a fake lock screen to steal credentials.
US citizen faces felony charges for deleting phone data at the border
A US citizen is facing serious felony charges after deleting data from his phone during a border search, a case that has sparked wide debate over digital rights at US borders.
Dev
Simon Willison's llm tool gets patched after OpenAI library change
llm 0.32.1 and llm-openrouter 0.7 ship fixes for installation breakage caused by OpenAI's Python library dropping its httpx dependency, while the openrouter plugin now surfaces model reasoning traces.
Claudette makes Claude stop writing like a BuzzFeed article
A new open source tool called Claudette adjusts Claude's prompting to cut the inflated, BuzzFeed-style writing tics and produce more direct answers.
Web
Kagi adds setting to filter paywalled links from search results
Search engine Kagi has added a setting that removes paywalled links from search results, a feature that drew heavy attention on Hacker News.