AI news of 15 August 2026

Anthropic explains how Claude’s invisible text watermarks will work

Anthropic details technical implementation of watermarking system for Claude outputs.

Anthropic has clarified how it's planning to apply invisible watermarks to Claude-generated text in order to comply with Europe's AI transparency rules. On Friday, Anthropic announced that Claude's text marking system is "a version of the SynthID-Text approach" - an open-source watermarking technology developed by Google DeepMind that creates detectable patterns using wording probabilities. This watermarking feature, alongside C2PA support for Claude-processed images, is being introduced to meet Anthropic's obligations under the European Union's AI Act, which requires synthetic audio, image, video, and text to include machine-readable marks …

The Verge 1mo ago also at TechCrunch

LLM evaluation

An eval harness found what qualitative review couldn't: AI models are most confident when wrong

Research finds LLM evaluation harness detects overconfidence that manual review missed.

VentureBeat 1mo ago

Grok, child safety

Woman claims her stepfather used Grok to transform childhood photo into explicit imagery Opens at the publisher in a new tab

Woman alleges Grok AI tool was used to create explicit imagery from her childhood photo.

TechCrunch 1mo ago

GLM-5.3, language models

GLM-5.3 is here with advanced cyber capabilities — and reportedly already found a 'serious vulnerability' in Cursor

Chinese startup releases GLM-5.3 language model with improved coding capabilities and discovers security flaw.

VentureBeat 1mo ago

AI in drug discovery

AI in drug discovery – what it is, where we stand and the path forward Opens at the publisher in a new tab

Nature review of AI applications in drug discovery, current state, and future research directions.

Hacker News 1mo ago

AI vision benchmarks

New benchmark confirms AI models still perform poorly at visual perception

Moonshot AI's PerceptionBench shows multimodal AI models score below 60% on visual perception tasks.

The Decoder 1mo ago

AI security, prompt injection

Plaintiff hid invisible AI instructions in court filings to secretly influence automated review

Connecticut plaintiff embedded hidden AI prompts in court documents to manipulate automated review systems.

The Decoder 1mo ago

SpaceX, Cursor

SpaceX officially closes its Cursor acquisition Opens at the publisher in a new tab

SpaceX completes acquisition of AI coding assistant startup Cursor.

TechCrunch 1mo ago

AI reasoning

AI Isn't Outthinking Mathematicians. It's Out-Remembering Them Opens at the publisher in a new tab

Analysis argues AI solves math through pattern memory rather than reasoning.

Hacker News 1mo ago

LLM tools

CORS Chat

Developer built a web UI tool to test large language models running locally on Mac and NVIDIA hardware.

Simon Willison 1mo ago

Anthropic explains how Claude’s invisible text watermarks will work

An eval harness found what qualitative review couldn't: AI models are most confident when wrong

Woman claims her stepfather used Grok to transform childhood photo into explicit imagery Opens at the publisher in a new tab

GLM-5.3 is here with advanced cyber capabilities — and reportedly already found a 'serious vulnerability' in Cursor

AI in drug discovery – what it is, where we stand and the path forward Opens at the publisher in a new tab

New benchmark confirms AI models still perform poorly at visual perception

Plaintiff hid invisible AI instructions in court filings to secretly influence automated review

SpaceX officially closes its Cursor acquisition Opens at the publisher in a new tab

AI Isn't Outthinking Mathematicians. It's Out-Remembering Them Opens at the publisher in a new tab

CORS Chat

Latest AI News

Choose sides Choose whose side you're on

Install it as an app

Ready for more?

Spread the word

PWA successfully installed.
Take that, Apple! 😉