Artificial Intelligence

Anthropic introduced Claude Fable 5.2 and Mythos 5.2 on November 9, 2026, with lower prices, an optional fast mode, and performance jumps in cybersecurity, biology, and robotics. We cover what changes from version 5.1 and what the published benchmarks show.

Anthropic included an unusual admission in its S1 filing with the SEC: its own pace of progress might be crossing the dramatic acceleration threshold defined by Anthropic’s RSP. A METR report backs this reading with data from the AECI index, though much of the evidence remains redacted.

The PHP2Shell vulnerability allowed remote code execution on NASA and Wikipedia sites, stemming from a memory bug ignored since 2024. An audit by Calif, funded by OpenAI, found the exploit chain in two weeks with AI’s help, and the PHP Foundation released a patch before any attacker could exploit it.

A developer documented how Postman installs its plugin in Codex and Claude Code without permission, and how the installer itself approves the startup hook that a human should review. The case, logged in GitHub issue #14370, exposes how easy it is to bypass consent in the code-agent plugin ecosystem.

Researchers at Parameter found a ChatGPT vulnerability that allowed attackers to take over any account just by opening a shared link. OpenAI fixed the flaw on October 26, 2026, and classified it as P1 severity.

Cadence, the giant behind semiconductor design software, joined the Genesis Mission consortium, the US program with more than $5 billion in funding to accelerate science with artificial intelligence. The company contributes Cerebrus, its reinforcement-learning-based tool, to optimize chip design.

ChipAgents, the startup founded by William Wang, applies agentic AI for chips to semiconductor verification. In an interview with Pulse 2.0, Wang explains how his agents generate and fix testbenches without constant manual intervention.