New: I've started Guilds, essays and a story about what gives life purpose once AI does more of the work. Essay 1 starts with London's weavers in 1130. guildsfutures.substack.com
1
1
97
A small UK gas plant went offline for four days in July after a cyber incident. New baseline cyber rules for Ofgem licensees apply by the end of 2030. Industry's top extra ask: IT/OT segmentation (18 of 49 replies). Small remote site: map, segment or lock remote access first?
1
25
Thomas Gere retweeted
Today, we've published our 2026 annual review: The State of the Cyber Nation. The UK faces a moment of choice. The future of cyber resilience is not set. Read the report now: ncsc.gov.uk/collection/annua… @GCHQ @DanJarvisMBE
16
77
19
214
18,670
1/6 In 1130, a man called Robert, son of Levestan, paid £16 into the King's Exchequer on behalf of London's weavers. It's the oldest record of a London guild. A thread on why I think AI will make us ask what guilds were for all over again. 🧵
1
43
5/6 If nobody needs a junior to do the routine work, who trains the next senior? That's the apprenticeship problem, and it's what guilds were built to solve.
1
6
6/6 My guess: guilds come back, not in their old form. Open entry, standing earned by contribution, and reputation you have to renew. Essay 1 of Guilds: guildsfutures.substack.com/p…
5
Stanford (Aug 2026): US jobs for 22 to 25s in highly AI-exposed roles are 19% behind less-exposed peers, mostly fewer hires. Early signal, not proof. If juniors stop doing routine work, who trains the next senior? Guilds solved that once. guildsfutures.substack.com
1
41
Schneider Electric, NVIDIA and HPE have launched AI agents that run at the plant edge and can act, not just advise: fix faults, respond to threats, adjust energy use. An agent that can act is a new privileged user on your OT network. Is the network ready?
2
1
72
My view: agents need reliable, segmented, logged site networks, with people approving critical actions. Licensed private 5G with standards-based interoperability fits. No savings or security results published yet. Schneider: globenewswire.com/news-relea…
40
The gap between frontier and open-weight AI is now about five weeks on one index (Kimi K3.1 Pro, weights due soon). That gap was defenders' lead time. Once weights are out, guardrails can be obliterated in minutes. For OT, there's no warning period left.
Based on this, Kimi K3.1 Pro is ~5 weeks behind the US frontier. That's the shortest the open / closed AI model gap has ever been.
1
1
57
"Obliteration" (abliteration in the research) deletes the refusal direction inside an open model, with no retraining. No provider monitoring, no kill switch. The capability stays.
1
19
Medieval guilds gave people three things: a craft, a standard to aim for, and peers who knew good work. As AI takes more of the routine, I think we'll want all three back. Essay 1 of Guilds: guildsfutures.substack.com/p…
25
Manufacturing ransomware victims: up to 39.7% more year on year, Jan to Jul 2026. Europe: up 85.4%. Attackers don't need to touch a controller to stop a plant. Hitting the IT behind it can be enough. If IT went dark today, could your line keep running?
1
1
58
Thomas Gere retweeted
AI is redefining cyber security best practice. That's why we have added a new chapter to our NCSC research problem book: 'AI in cyber security problems’. To learn more and contribute to the discussion⬇️ ncsc.gov.uk/blogs/introducin…
10
65
3
222
12,998
Thanks @DigiCatapult for hosting the Scaling Private 5G Adoption focus group. Good to meet peers and learn. Their new blueprint is worth a read: - standards-based, interoperable design - sites run on if cloud links drop - security built in from day one digicatapult.org.uk/blogs/po…
32
AI found flaws in PHP. Attackers moved fast. The UK took 32 government services offline, the NCSC sent 50,000 warnings, and Ofcom told telcos to act. How fast could your site patch or isolate a critical flaw? This week on AI CyberWatch: aicyberwatch.org/news
32