We tested Claude, GPT-5.2, and Gemini as scientific paper reviewers on @KurateOrg. When scoring arXiv papers on impact (1–10), GPT clusters around 7–8 and Gemini is similar. Claude Opus 4.6 uses the full range, making it a far better discriminator. See the distributions 👇

Anthropic launched Project Glasswing in April 2026 to tackle a growing cybersecurity problem: software has more security flaws than humans can realistically find and fix. The company gave about 50 organizations early access to a new AI model called Claude Mythos Preview, designed specifically to search for vulnerabilities in code. The results exceeded expectations. >In just one month, the system found more than 10,000 high or critical security vulnerabilities. >Scanning over 1,000 open-source projects uncovered 23,019 total issues >6,202 were rated high or critical severity. >Independent reviewers confirmed that about 90.6% of the findings were real problems. Several organizations reported major discoveries: >Cloudflare identified around 2,000 bugs, including 400 high or critical ones. >Mozilla fixed 271 vulnerabilities in Firefox version 150, more than 10 times the number fixed in the previous release. >The project also rediscovered old flaws, including a 27-year-old vulnerability in OpenBSD and a 16-year-old issue in FFmpeg.

Qureate CEO Yuji Usuda, head of the Japanese game company Qureate, recently explained why the studio’s “gentlemen’s games” almost always feature busty female characters. In a interview with Game*Spark, Usuda said that portraying adult women with small or flat chests has become “highly risky” because platforms and reviewers may misinterpret petite body types as underage characters. Even clearly adult characters can create problems for releases on platforms like Steam or Nintendo Switch. He said the studio would like to include more body variety, but avoiding release issues has become a priority. “If we took on risky depictions and ended up blocked from releasing the game at all, it would be pointless,” he explained. Usuda also mentioned that Qureate already has to make platform-specific changes to avoid censorship. In Bunny Garden, for example, facial “lotion” was changed to green slime and milk effects were made invisible in some versions.

A Steam reviewer recently said they were “done with” Destiny 2 after putting more than 10,000 hours into the game and leaving a negative review. The funny part is the guy went on to play another 160 hours after posting the negative review.

I had Opus 4.8 in Claude Code write a sophisticated, if minor, academic paper from a archive of hundreds of de-identified research files from years ago I had to use GPT-5.5 Pro as a reviewer, it spotted one major error & some minor points. Opus corrected

This theorem (left) means, the only way you can make proofs for two different things in the same position in the same Merkle tree, is by breaking the underlying hash function. As a reviewer, you don't have to verify how Merkle branches are implemented or how the theorem is proven (right), you just have to verify what the theorem says, and that Lean verifies it. And the beautiful thing is that you can even write live production code (including eg. CLI tools) directly in Lean.

In Claude Code, the new /ultrareview command runs a dedicated review session that reads through your changes and flags what a careful reviewer would catch. We've also extended auto mode to Max users, so longer tasks run with fewer interruptions.