Thursday’s tech news November 5 2026 is a tale of two AIs: one that breaks into systems, and one that helps scientists understand life itself. OpenAI turned its agents loose on the Linux kernel and came back with nearly a thousand working root exploits, while Anthropic’s Claude helped crack a biology puzzle that has stumped researchers for four decades.
Key Takeaways
- OpenAI’s security agents found, exploited, and drafted patches for 948 distinct Linux kernel flaws in one run costing about $1.2 million.
- Anthropic’s Claude helped map the first 3D structure of a vault particle from bacterial life, compressing days of lab work into about 90 minutes.
- Apple cut its engineers’ daily AI spending allowance from $300 to $200, with deeper cuts reportedly on the way.
- Tencent’s Hy4 gained vision and computer use, but the downloadable model weights are still pending.
- Meta acquired Conductor, an AI coding startup, the same day OpenAI moved to buy Pierre.
1. OpenAI’s AI Agents Found 948 Ways to Take Over Linux
According to OpenAI’s research post “A thousand fewer paths to root,” the company’s Cyber team built an autonomous pipeline that hunts for security flaws, proves they are real, builds working exploits, and drafts patches, with humans reviewing the results first.
The mid-October run sifted through 2,308 candidate findings, reproduced 1,724 as genuine vulnerabilities, and produced working exploits for 948 distinct local privilege-escalation flaws — the kind that let an ordinary logged-in user grab full root control. The whole run cost $1,213,430 in API credits, or roughly $1,279 per flaw. The patches are candidates awaiting maintainer review, not fixes already shipped. Still, OpenAI’s guidance is blunt: treat any unpatched system with a low-privileged account as equivalent to root access for a well-resourced attacker. For more on the security side of AI, see our tech news roundup from November 4, which covered CrowdStrike’s AI Kill Switch.
2. Claude Helps Solve a 40-Year-Old Biology Mystery
Anthropic’s biology lab announced it has mapped the first three-dimensional structure of a vault particle found in prokaryotic cells. Vaults are giant hollow barrels of protein that have puzzled scientists since their discovery in 1986. Nobody knows what they actually do.
According to Anthropic, Claude helped at nearly every step, from searching roughly 15 billion predicted proteins to processing the microscope data. A purification that conventionally takes three days took about 90 minutes, and a first structural model was ready in 34 hours — a job the team estimates would normally have taken weeks. One of Claude’s protocols even cost the project two weeks, when a growth additive it suggested blocked DNA from entering cells. AI does not do science alone, but a skilled human team with an AI assistant can move dramatically faster — as long as someone is checking the work. Our November 3 roundup covered Anthropic’s reported $895 billion compute bill.
3. Apple Starts Rationing Its Own Employees’ AI Use
Apple is cutting the daily amount its engineering teams can spend on AI, Bloomberg reported on Wednesday. In October, the company lowered the daily AI spending cap for its engineers from $300 to $200, with more cuts reportedly in the works — some allowances falling to $100 a day, and a limit as low as $25 weighed for staff outside engineering.
The budgets pay for AI tokens, the units models like ChatGPT, Claude, and Gemini process, and employees who need more can reportedly request exemptions from senior leadership. Earlier this year, many companies pushed staff to use AI tools as much as possible. Now the bills are coming due, and even Apple is watching the meter. Our review of Apple’s touchscreen MacBook Pro covers the hardware side of Apple’s big week.
4. Tencent’s Hy4 Gets Eyes and Hands — but the Weights Are Still Coming
Tencent released the full version of Hy4 on November 4, adding vision and computer use to the model it previewed in August. The 770-billion-parameter model can now read images and charts and operate command-line and graphical interfaces, including 3D and CAD tools.
There is a catch. The August preview shipped with its weights on day one under an open license, but the full release’s weights are not out yet. Tencent says it plans to publish them, along with a system card, in the coming days. Until then, Hy4 is available through APIs only — via Tencent Cloud, OpenRouter, and Tencent’s own apps. The weights will be the real test of how open this release turns out to be.
5. Meta Buys Conductor as the Coding-Agent Land Grab Heats Up
Meta has acquired Conductor, an AI coding startup that lets developers run multiple coding agents in parallel workspaces, each working on a different part of a project at the same time. According to TechCrunch, the deal was first reported by Axios, and founder and CEO Charlie Holtz announced it on the company’s blog. Meta confirmed the acquisition but declined to share financial terms.
Conductor does not build its own models — it coordinates agents from labs like Anthropic and OpenAI. If you use Conductor’s cloud service, mark your calendar: it shuts down on November 20, after which the company will begin deleting user cloud data. The same day, OpenAI said it will acquire Pierre Computer Company, whose code-navigation technology already powers parts of its Codex coding agent. Two deals in one day tells you where the competition has moved: the models are table stakes now, and the fight is over the tools that organize them.
The Takeaway
Today’s stories all point the same way: AI is no longer just a thing you chat with. It finds security flaws and writes the patches. It is expensive enough that Apple is rationing it, powerful enough that Tencent is racing to open it up, and valuable enough that Meta and OpenAI are buying the companies that organize it. The age of the AI assistant is giving way to the age of the AI workforce.
Check back tomorrow for the next roundup.
FAQ
How many Linux kernel flaws did OpenAI’s AI agents find?
OpenAI’s autonomous security pipeline reproduced 1,724 real vulnerabilities in the Linux kernel and produced working exploits for 948 distinct local privilege-escalation flaws, each with a draft patch. The mid-October 2026 run cost about $1.2 million in API credits.
What is a vault particle, and why does Claude’s discovery matter?
Vault particles are large, hollow protein barrels found in cells across the tree of life, and their function has been unknown since their discovery in 1986. Claude helped Anthropic’s biology lab map the first 3D structure of a vault from bacterial life, compressing a three-day purification into about 90 minutes.
Why is Apple cutting its employees’ AI budgets?
Apple lowered its engineers’ daily AI spending cap from $300 to $200 in October 2026, with further cuts reportedly planned, as part of a broader cost-control push. The budgets cover AI tokens used across tools like ChatGPT, Claude, and Gemini.
When will Tencent Hy4’s open weights be released?
Tencent says it plans to release the full Hy4 model’s weights and a system card in the coming days, but no date has been given. The model is currently available through APIs only, via Tencent Cloud, OpenRouter, and Tencent’s apps.
Sources
- OpenAI, “A thousand fewer paths to root” (November 4, 2026) — reported via GBHackers
- Anthropic, “Investigating vault particles with Claude” — https://www.anthropic.com/research/investigating-vault-particles-with-claude
- TechCrunch, “Apple is reportedly cutting budgets for its internal AI usage” (November 4, 2026) — https://techcrunch.com/2026/11/04/apple-is-reportedly-cutting-budgets-for-its-internal-ai-usage/
- ExplainX, “Tencent Hy4 Release: 770B, Computer Use, 12 Benchmarks” — https://explainx.ai/blog/tencent-hy4-official-release-770b-vision-computer-use-terminal-bench-44-4-weights-planned-2026
- TechCrunch, “Meta acquires AI coding startup Conductor” (November 4, 2026) — https://techcrunch.com/2026/11/04/meta-acquires-ai-coding-startup-conductor/
