In partnership with
Hello from The AI Night,
Today in AI:
Anthropic Announces Project Glasswing for AI Cybersecurity
Z AI Launches GLM-5.1 for Long-Horizon Coding Tasks
Tesla Rolls Out FSD v14.3 with 20% Faster Reaction Time
Here's the deal: Anthropic announces Project Glasswing, a cybersecurity initiative backed by AWS, Apple, Google, Microsoft, NVIDIA, and eight other major partners. At its core is Claude Mythos Preview, an unreleased frontier model that has already found thousands of zero-day vulnerabilities across every major operating system and web browser.
The Breakdown:
Mythos Preview discovered flaws that survived decades of human review, including a 27 year old bug in OpenBSD and a 16 year old vulnerability in FFmpeg that automated tools missed after five million test runs.
The model scores 93.9% on SWE-bench Verified versus 80.8% for Opus 4.6, and 83.1% on CyberGym versus 66.6%.
Anthropic is committing $100M in usage credits plus $4M in donations to open-source security organizations.
Over 40 additional organizations maintaining critical infrastructure will get access to scan and patch their systems.
Mythos Preview will not be made generally available. Pricing for partners after credits; $25/$125 per million input/output tokens.
The bigger picture: This signals AI capability has crossed a threshold where vulnerability discovery at scale is no longer limited to elite human researchers, forcing the entire industry to rethink defensive security timelines.
Here's the deal: Z AI launches GLM-5.1, an open-source (MIT License) flagship model designed for long-horizon agentic tasks. The core claim; unlike predecessors that exhaust their strategies early, GLM-5.1 sustains meaningful optimization over hundreds of iterations and thousands of tool calls.
The Breakdown:
On a vector database optimization task, GLM-5.1 reached 21.5k QPS over 600+ iterations, roughly 6x the best single-session result (3,547 QPS by Claude Opus 4.6 in 50 turns).
On KernelBench Level 3 GPU kernel optimization, GLM-5.1 hit 3.6x speedup, though Claude Opus 4.6 led at 4.2x.
On SWE-Bench Pro, GLM-5.1 scored 58.4%, edging out Claude Opus 4.6 (57.3%) and GPT-5.4 (57.7%).
In an 8-hour web app build test, the model iteratively constructed a full browser-based Linux desktop environment without intermediate guidance.
Weights are available on HuggingFace and ModelScope. Compatible with Claude Code, vLLM, and SGLang.
The bigger picture: GLM-5.1 reframes model capability as not just first-pass accuracy but sustained usefulness over time. For builders running complex agentic workflows, longer productive horizons directly translate to better outcomes without human intervention.
Here's the deal: Tesla has started rolling out Full Self-Driving (Supervised) v14.3, a significant update to its autonomous driving software. The release targets faster response, better edge-case handling, and improved training infrastructure across the board.
The Breakdown:
AI compiler and runtime were rewritten from the ground up using MLIR, delivering 20% faster reaction time and speeding up internal model iteration.
Reinforcement learning training stage was upgraded, improving performance across a wide range of driving scenarios.
Vision encoder upgrade strengthens 3D geometry understanding, expands traffic sign recognition, and improves handling of rare and low-visibility conditions.
Hard RL examples sourced from the Tesla fleet now train the system on complex intersections, unusual objects in the vehicle path, and small animal detection.
Parking got more decisive spot selection and a new map pin with a (P) icon for predicted parking locations.
Upcoming additions include pothole avoidance, expanded reasoning beyond destination handling, and improved driver monitoring with better eye gaze tracking.
The bigger picture: The MLIR rewrite signals Tesla is investing in foundational inference infrastructure, not just model quality. Faster reaction time and fleet-sourced hard examples suggest a tightening loop between real-world data collection and model improvement, which is the core scaling advantage Tesla has been building toward.
Turn AI into Your Income Engine
Ready to transform artificial intelligence from a buzzword into your personal revenue generator?
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
What else you need to know:
SpaceX's Colossus 2 AI cluster is now training seven models simultaneously, ranging from Imagine V2 to a 10T-parameter model, with Elon Musk noting progress gaps remain.
Intel announced it is joining the Terafab project alongside SpaceX, xAI, and Tesla to refactor silicon fab technology, aiming to produce 1 terawatt per year of compute for AI and robotics.
X launched a built-in photo editor in its post composer with drawing, text overlay, blur redaction, and a Grok-powered "edit with words" feature, now live on iOS with Android coming soon.
OpenAI is retiring six older Codex models, including gpt-5, gpt-5.1, and gpt-5.2-codex variants, on April 14 when Codex transitions to ChatGPT account sign-in.
Cursor launched cloud agents that run in remote sandboxes and can be controlled from web, mobile, Slack, or GitHub, producing merge-ready pull requests with visual artifacts for validation.
That’s it for today’s edition of The AI Night.
Our goal is to cut through the noise, surface what actually changed, and explain why it matters.
3 ways to support us:
Forward this to your AI-curious friend → https://www.theainight.com
Sponsor The AI Night and reach 500+ AI builders daily → passionfroot.me/theainight
Reply to this email — I read every response