- Bay Area Times
- Posts
- Anthropic launches Project Glasswing using Claude Mythos, saying it identified thousands of critical bugs
Anthropic launches Project Glasswing using Claude Mythos, saying it identified thousands of critical bugs
Bay Area Times
April 08, 2026
Brought to you by:
Top stories today:
- Anthropic launches Project Glasswing using Claude Mythos
- Z.ai releases GLM-5.1 open-weight model, rivaling GPT-5.4, Opus 4.6
- Elon amends OpenAI lawsuit, seeks damages for nonprofit
- Perplexity ARR tops $450M in Mar., +50% on agent launch
- U.S., Iran agree to two-week ceasefire hours before Trump deadline
- Google adds vertical tabs, new reading mode to Chrome to match rivals
0. Data and calendar
All times are ET.
1. Anthropic launches Project Glasswing using Claude Mythos, saying it identified thousands of critical bugs
Claude Mythos Preview has shown significantly stronger agentic coding and reasoning than Opus 4.6 in early benchmarks, Anthropic said:
The unreleased model has identified thousands of zero-day vulnerabilities across major OS, web browsers, and critical software, the company said.
A 27-year-old OpenBSD bug and a 16-year-old FFmpeg flaw are among them.
Mythos scored 82% on Terminal-Bench 2.0, vs. GPT-5.4’s 75.1% and Opus 4.6’s 65.4%.
“We did not explicitly train Mythos Preview to have these capabilities,” Anthropic said.
40+ organizations will get Mythos Preview, the company said.
No plans to make Mythos generally available.
The project brings together 12 tech companies to secure critical software
Anthropic has committed up to $100M in usage credits, along with $4M in donations to open-source security groups.
Want to know what world-class talent actually costs in 2026?
Athyna's Salary Report breaks down real salary data across AI, Tech, Data, Design, and more—so you can see exactly where the savings are.
The numbers might surprise you.
*Disclaimer: We have equity in Athyna.
3. Z.ai releases GLM-5.1 open-weight AI model, rivaling GPT-5.4, Claude Opus 4.6
GLM-5.1 has topped open-weight models on the Artificial Analysis Intelligence Index with a modest gain over GLM-5:
The model has 744B parameters (40B active) and a 200K context window, matching GLM-5.
Licensed under MIT.
GLM-5.1 is priced at $1.4/$4.4 per million input/output tokens, vs. GLM-5’s $1/$3.2, MiniMax-M2.7’s $0.3/$1.2, GPT-5.4’s $2.5/$15, and Opus 4.6’s $5/$25.
Available via Z.ai’s API and third-party providers including DeepInfra, Fireworks AI, and GMI Cloud.
GLM-5.1 has the lowest hallucination rate among open-weight models on AA-Omniscience
GLM-5.1 is slightly less token-efficient than GLM-5
4. Elon amends OpenAI lawsuit, seeks damages for nonprofit, Altman’s removal from board
OpenAI called the move “nothing more than a harassment campaign” in a post on X:
