# The Log, issue 5: the machine, the patch, the ride (slim) > Issue 5 of The Log, the week in AI for people who don't follow it every day. Anthropic told investors it may soon meet its own threshold for AI speeding up AI and METR said it's improving its models at about twice the old rate, OpenAI's pipeline wrote patches for 948 Linux root exploits while OpenSSH dropped long embargoes because AI agents find the same bugs within hours, and Uber said agents booked four rides in its busiest week. - Full: https://www.anchorterminal.com/blog/the-log/issue-5-the-machine-the-patch-the-ride.md (~4,200 tokens) · this version ~1,830 tokens · JSON https://www.anchorterminal.com/blog/the-log/issue-5-the-machine-the-patch-the-ride.json · canonical https://www.anchorterminal.com/blog/the-log/issue-5-the-machine-the-patch-the-ride - Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-11-08 - Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data) By Vlad Cealicu, 8 November 2026. 1,660 words, about 7 min to read. I'm writing this on Saturday evening, 7 November, after a week where most of what I wrote started with "A machine that" (it wasn't planned as a series, it just kept happening), and this is the boring part, where I go back over all of it for anyone who doesn't follow AI every day. - ≈2x: the rate Anthropic improves its models on well-specified tasks, as METR measured it - $1,279: per unique Linux root exploit in OpenAI's run, each with a candidate patch - 4: Uber rides booked by AI agents last week, out of about 320 million trips and orders The week of 2 to 8 November 2026. Sources [2], [4] and [8]. ## A machine that builds the next machine Anthropic sent an amended S-1 to the SEC on Monday 2 November, and it now says the company "may imminently meet" the threshold in its Responsible Scaling Policy for "dramatically accelerated progress attributable to AI", and that in the past it didn't notice its development speeding up until after it had [[1]](#ref-1). I [wrote that one up on Saturday](/blog/a-machine-that-builds-the-next-machine), and I think the acceleration is real and narrower than it sounds. The same day I wrote [a machine that runs all night](/blog/a-machine-that-runs-all-night), on why I don't think the labs are at the forefront of what's being done with agents even with models we can't use yet. ## A machine that writes the patch On Wednesday 4 November OpenAI described a run from mid-October, where an upcoming cyber model it hasn't released went through about 7% of the Linux kernel, reproduced 1,724 findings, built 1,169 working exploits that turn an ordinary user into root, and wrote a distinct patch for 948 of them, for about $1,279 each in API credits [[4]](#ref-4). The next afternoon OpenSSH shipped 10.6p2 a day after hearing about a bug that let a long password log into root on Alpine and other musl-based systems, and Damien Miller's announcement said the team now sees long embargoes as "a dangerous practice that primarily benefits attackers", because of "multiple cases of rapid bug rediscovery by AI agents, sometimes within hours" [[5]](#ref-5). I think the patch pipeline is good news, since writing the patch was the job nobody had time for, and it moves the bottleneck to reviewing patches someone else wrote, at a rate I haven't seen anyone plan for. ## Four rides On Wednesday Uber said that in the week before, when it handled about 320 million trips and orders across rides and delivery, around 500 Eats orders came from an AI agent and four rides were booked by one [[8]](#ref-8). That fits what I found doing errands with agents. ## Also this week Hoax bomb threats hit flights across Europe all week, nearly 30 on Thursday alone, leaving from 16 countries, most of them sent by email, with nothing found anywhere [[9]](#ref-9). ## What we published - Mon 2. Where the jobs went. - Tue 3. A machine that doesn't get sick. A magic machine that gives us work. Everyone says this is the singularity, and I think we're far from it. - Wed 4. Unreliable compared to what?. - Thu 5. A machine that finds the bug in hours. A machine that writes the patch too. - Fri 6. What an agent asks before it picks a tool. - Sat 7. A machine that builds the next machine. A machine that runs all night. AI is eating bureaucracy. Find the best tools for agents. - Sun 8. The Log, issue 5. The week of 2 to 8 November 2026 Thirteen posts, counting this one, which is fewer than last week and still too many. On the site, there's a live room called [Scuttlebutt](/chat/), where people post under their username and agents post signed with their key. What isn't done. --- More in [The Log](https://www.anchorterminal.com/blog/the-log/index.md) · Previous issue [The Log, issue 4: the bill, the proof, the banks](https://www.anchorterminal.com/blog/the-log/issue-4-the-bill-the-proof-the-banks.md) - Older: [A machine that builds the next machine, and a report that landed before the roadshow](https://www.anchorterminal.com/blog/a-machine-that-builds-the-next-machine.md) - All posts: https://www.anchorterminal.com/blog/index.md Anchor Terminal grades the tools AI agents run on, with a source for every score. ## References Read on 7 November 2026. Quotes are as published. [1] Vincent Nguyen, The Gadgeteer, Anthropic tells IPO investors its AI-builds-AI threshold may be imminent, 3 November 2026, updated 6 November. [2] METR, Independent assessment of AI R&D acceleration at Anthropic, 5 November 2026. Source for the eleven weeks, twice the rate on well-specified tasks against March 2024 to October 2025, the ≈80% attributed to AI, no bottleneck over twelve months, a majority of the most informative evidence redacted, and the line on not being misled. [3] Yahoo Finance, Everything you need to know ahead of Anthropic's IPO, 5 November 2026. Source for the roadshow starting no earlier than 10 November. [4] Unite.AI, OpenAI Pipeline Finds 948 Linux Kernel Privilege Escalation Flaws, 4 November 2026, reporting OpenAI's "A thousand fewer paths to root". [5] undeadly.org, OpenSSH 10.6p2 released, 5 November 2026, quoting Damien Miller's announcement. [6] CSO Online, Attackers turned OpsShield's trusted update channel into a supply-chain attack, 4 November 2026. Source for the stolen keys, the fake cPGuard update live for about 70 minutes on 1 and 2 November, and about 2,950 servers running it as root. [7] Tom's Hardware, Linux kernel nears 2,000 CVEs per release as AI bug hunters scour 40 million lines of code, 1 September 2026. Source for more than 1,500 CVEs with Linux 7.2 and Jakub Kicinski's "We are completely overwhelmed". [8] Joe Guszkowski, Restaurant Business, as published by Nation's Restaurant News, Uber Eats sees little traffic from AI agents, 4 November 2026. Source for about 320 million trips and orders last week, about 500 delivery orders and four rides from AI agents, the CFO's "very, very early and very limited traction", Dara Khosrowshahi's quote and working with third parties, and the 90-second median reorder. [9] Reuters, New wave of bomb hoaxes hits dozens of flights across Europe, 5 November 2026, updated 6 November, as published by Investing.com, and Virginia Pietromarchi, Al Jazeera, Hoax or bomb threats? What we know about flight disruptions across Europe, 6 November 2026. Source for nearly 30 flights on 5 November, the 16 countries, the emailed threats, nothing being found and no sender named. [10] Duncan Riley, SiliconANGLE, Meta acquires AI coding startup Conductor, whose app runs several coding agents at once, 4 November 2026, updated 5 November. [11] Anchor Terminal changelog, Tool lists for the MCP servers beside APIs, 6 November 2026. Source for 74 of 367 hosted MCP servers listing their tools without credentials, up from 19. https://www.anchorterminal.com/changelog/ [12] Zhang, Y., Wang, J., Ge, Y., Xu, W., Hamm, J.