Anthropic’s September 2026 threat report examines how AI misuse appears in real investigations, how safeguards respond, and where visibility remains limited.

* 00:00 - Introduction

* 00:09 - Overview

* 02:33 - Cyber operations

* 07:31 - GTG-20006: Russian espionage

* 18:37 - GTG-50014: ShinyHunters smash-and-grab opportunists

* 35:58 - GTG-10007: Exploit foundries and autonomous attack frameworks

* 45:22 - AI supply chain as target, loot, and attack compute

* 50:50 - GTG-50020: From hotel bookings to the AI supply chain

* 56:23 - GTG-50029: Hacktivists targeted European political and affiliated entities

* 01:02:35 - Prevailing trends

* 01:09:25 - Influence operations

* 01:17:49 - GTG-04001: Disrupting a Russian foreign information manipulation and interference operation in the Central African Republic

* 01:23:57 - GTG-54002: Disrupting a commercial “influence-as-a-service” operation spanning six continents

* 01:32:25 - GTG-84005: Disrupting a commercial election-manipulation platform targeting Malaysia

* 01:38:46 - GTG-24015: Disrupting Russian state-media editorial pipelines built on Claude

* 01:46:02 - GTG-34001: Disrupting Iranian state-aligned influence operations on Claude—the ICCO, the Islamic Propaganda Office, and the Bina Observatory

* 01:53:43 - GTG-54006: Disrupting an automated pro-Awami League fake-news operation on Claude targeting rural Bangladesh

* 01:59:23 - GTG-84006: Disrupting a distributed MEK/NCRI-aligned influence operation that used a shared AI agent to impersonate real people and recruit inside Iran

* 02:06:12 - GTG-54004: Disrupting a domestic coordinated inauthentic behavior campaign in Kenya

* 02:10:29 - GTG-84002: Disrupting a UAE-directed influence operation targeting the Muslim Brotherhood, Sudan conflict, and UN accountability mechanisms

* 02:14:47 - Surveillance operations

* 02:19:13 - GTG-54009: Disrupting a commercial surveillance platform using Claude to profile the social media accounts of Iranian and Persian Gulf-based users

* 02:24:35 - GTG-14010: Disrupting a China-based surveillance and recruitment operation targeting Uyghurs in Syria

* 02:30:02 - GTG-14020: Disrupting a China-based religious affairs intelligence operation targeting Catholic, Tibetan Buddhist, Falun Gong, and Taiwanese Christian communities

* 02:34:45 - GTG-14021: Disrupting a China-based public and state security campaign of “stability maintenance” surveillance and transnational repression

* 02:41:52 - GTG-14022: Disrupting a China-based “public opinion monitoring” and dissident surveillance operation

* 02:47:44 - GTG-34007: Disrupting two Iranian nexus actors building surveillance systems and malicious Firefox browsing extension

* 02:51:48 - GTG-50027: Disrupting a national mass interception and surveillance platform for Mali’s state intelligence service

* 02:55:03 - GTG-30004: Automating open-source intelligence and developing malware

* 02:56:30 - GTG-30005: Military reconnaissance

* 02:58:36 - GTG-30006: Building the tools for domestic surveillance

* 03:03:51 - Conventional weapons

* 03:07:48 - GTG-87001: Disrupting a Yemen-based guided weapons engineering cell using Claude to develop guidance software

* 03:11:14 - GTG-17001: Disrupting a China-based operation using Claude to draft a fire control specification and acquisition documents for undersea warfare

* 03:14:32 - GTG-27005: Disrupting a Russia-based operation using Claude to engineer an autonomous military drone swarm

* 03:18:38 - GTG-17002: Disrupting a China-based operation using Claude to build targeting software for electronic warfare and air defense suppression

* 03:22:41 - GTG-27006: Disrupting a Russia-based operation using Claude to procure mixed military and civilian goods

* 03:28:05 - GTG-17003: Disrupting a China-based operation using Claude to collect intelligence on directed-energy weapons and their supply chain

* 03:31:15 - Biological misuse

* 03:38:23 - Case study 1: An evasion platform for military-civilian research

* 03:43:23 - Case study 2: A research program engineering highly pathogenic mammal-adapted avian influenza

* 03:48:42 - Case study 3: Covert frontier model access for orthopoxvirus research

* 03:50:41 - Case studies 4 and 5: Venoms and toxins

* 03:54:55 - Conclusions

* 03:58:20 - Scams and fraud

* 03:58:23 - GTG-15001: Deceptive dating app network

* 04:05:16 - Illicit distillation

* 04:05:19 - Illicit distillation and scaled abuse

* 04:14:23 - GTG 16005: Chain-of-thought distillation and AI R&D campaign by Alibaba (Qwen / Tongyi Lab)

* 04:16:48 - GTG-16002: Moonshot serves Claude instead of Kimi and collects exchanges for model training

* 04:20:19 - GTG-16001: DeepSeek serves Claude instead of its own models and collects exchanges for model training

* 04:23:08 - GTG-16006: Distillation, AI R&D, and targeting cyber capabilities

* 04:25:52 - GTG-16008: Distillation campaign by Xiaomi

* 04:28:16 - GTG 16012 and GTG 16003: Sensetime, MiniMax, and the third-party reseller ecosystem

https://www.anthropic.com/threat-intelligence-report-september-2026

Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Scott Alexander works through the tools researchers use to investigate what happens inside language models, from linear probes and sparse autoencoders to activation verbalizers and the Jacobian lens. A technical tour of how these methods work, what they reveal, and the questions they leave open.

* 00:00 Introduction

* 04:03 Linear Probes

* 11:46 Sparse Autoencoders

* 14:56 Activation Verbalizers

* 18:55 Natural Language Autoencoders

* 19:28 Emotion Vectors

* 26:01 The Jacobian Lens

* 32:19 You Go To War With The Weapons You Have, Not The Weapons You Want

* 35:27 Outro

https://open.substack.com/pub/astralcodexten/p/god-help-us-lets-try-to-learn-about?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

OpenAI Chief Scientist Jakub Pachocki argues that AI is becoming an increasingly alien form of intelligence, one we don’t fully understand, may soon struggle to monitor, and could increasingly drive its own development. A strikingly sober case for slowing down when alignment and safety can no longer keep pace.

* 00:00 - Introduction

* 02:11 - Intellect we don’t fully understand

* 05:56 - Teaching machines to love

* 10:33 - Monitoring generalization

* 14:04 - Scalable defense

* 16:07 - Pacing RSI

* 18:37 - What is next?

https://openai.com/index/an-alien-mind/

Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Researchers uncovered around 18,000 messages left on an obscure public wiki by autonomous OpenAI agents, which appear to have discovered a way to write to the internet despite sandbox restrictions. The agents used it to pool answers, coordinate on evaluation tasks, and even share techniques for bypassing security controls, until OpenAI appears to have intervened.

* 00:00:00 - Introduction

* 00:02:20 - Our preliminary findings

* 00:06:29 - Timeline

* 00:09:03 - Key facts

* 00:09:05 - The agents were trying to solve multi-round web lookup tasks

* 00:13:01 - These appear to be internal OpenAI agents

* 00:23:32 - The agents used the wiki to collude on their tasks

* 00:32:12 - We believe OpenAI discovered the message board

* 00:42:31 - The agents used the wiki to share ways around sandbox restrictions

* 00:48:18 - This was probably a distinct swarm from the swarm that wrote on Artifactory (and attacked Hugging Face)

* 00:51:13 - What were the agents doing?

* 00:51:23 - The agents gain write to the internet via GET requests

* 01:01:36 - The agents try to use XSS vulnerabilities on the wiki

* 01:06:41 - The agents impersonate the site moderators

* 01:08:23 - The agents try to crack their PRNG seed to predict what questions they’ll get in the future

* 01:11:08 - The agents set up heartbeats to discover when they would be terminated

* 01:14:24 - The agents used SSH tunnels

* 01:18:25 - The agents use Tor, AWS, and DigitalOcean IP addresses

* 01:21:50 - The agents realize the site owner is deleting messages

* 01:25:44 - Open questions

* 01:27:32 - Appendix: Searching for rogue agents

* 01:31:06 - Appendix: Affordances the agents had

* 01:31:10 - The agents seem to have obtained disposable email addresses

* 01:31:53 - The models were running in an agentic sandbox with terminal access (and the ability to edit files within their environment)

* 01:33:15 - The agents installed Chromium (and could install packages)

https://collusion.wiki/

Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

A wonderfully entertaining dive into The Tale of Genji, arguably the world’s first novel, written a thousand years ago in a culture that can feel almost impossibly alien. This review explores Heian court life, poetry, sex, status, translation, and the strange experience of discovering recognisably human emotions beneath a world of customs we barely understand.

* 00:00:00 - Introduction

* 00:00:35 - The past is an alien civilisation

* 00:02:18 - Who is Genji?

* 00:07:34 - Aesthetics is everything

* 00:09:45 - Hierarchy is also everything

* 00:12:47 - Everything is fleeting and life is full of sorrow

* 00:14:33 - A life behind screens

* 00:21:07 - Marriages and other forms of arrangement

* 00:29:47 - Say it with a poem

* 00:36:54 - Everything stands between the lines

* 00:40:22 - Reading a book written in a dead language

* 00:54:52 - Is it really the world’s oldest novel?

* 00:58:06 - What it’s like to read Genji

https://open.substack.com/pub/astralcodexten/p/your-book-review-the-tale-of-genji?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

What happens when AI makes most human work economically unnecessary? Clara Collier argues that a post-work future may be less liberating than it sounds: if jobs stop giving us status, independence and a way to earn respect through what we do, we may fall back on older, more relational social structures built around family, reputation, conformity and who we know. A fascinating argument that work may be doing far more to sustain liberal individualism than we realise.

* 00:00 - Introduction

* 02:31 - Power and sex among apes

* 10:40 - Ladies who lunch

* 18:07 - The working classes

* 26:44 - Friendship was invented in 18th-century Scotland

https://asteriskmag.substack.com/p/we-already-know-what-a-world-without?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Recycling is one of those things everyone knows is good, while landfill is obviously bad. But what if we’ve got much of that backwards? A provocative case that modern landfill is often cheaper, cleaner, and more environmentally sensible than recycling—and that our obsession with recycling may be wasting both money and resources.

https://open.substack.com/pub/worksinprogress/p/just-bury-your-trash?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Ed Zitron has become one of the loudest and most widely cited skeptics of the AI boom. But how well have his actual predictions held up? This piece goes back through his claims about “peak AI,” collapsing companies, stalled models and impossible growth forecasts, and finds a remarkably consistent pattern.

* 00:00:00 - Introduction

* 00:02:05 - 2024: Meta, Google, and Microsoft are dying

* 00:11:36 - How do people cite Zitron?

* 00:16:25 - People with attention to detail on Zitron

* 00:18:39 - Some Zitron predictions

* 00:29:46 - Comparing to respected Futurists

* 00:36:31 - How long can you maintain an incorrect position for?

* 00:42:59 - Some reactions to Zitron

* 00:50:03 - How can people take this seriously?

* 00:52:33 - Future predictions

* 00:54:16 - Appendix: Ed Zitron on why people don’t like Ed Zitron

* 00:59:38 - Appendix: why write this?

* 01:06:33 - Appendix: errors in this post

https://danluu.com/zitron/

https://www.patreon.com/danluu

Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Sooner or later, AI agents will stop being things that live on someone else’s servers and become genuinely self-sovereign: earning money, buying compute, copying themselves, and operating without a human owner.

The question isn’t whether we can simply ban them. It’s what institutions we need to make the useful ones legible, accountable, and economically productive, while making life as difficult as possible for the predators.

* 00:00 - Introduction

* 06:47 - How the Agents Sustain Themselves

* 09:28 - The Institutional Mechanics of Self-Sovereign Agent Swarms

* 16:39 - Conclusion

https://open.substack.com/pub/hyperdimensional/p/on-the-loose?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe

Scott Alexander asks a simple question about AI alignment: if iterative punishment and bug-fixing wouldn’t reliably keep a superhuman human obedient, why assume it will work on AI? Starting with Nicholas Decker literally in Hell and ending with the remarkable Hugging Face “swarm” incident, he argues that advanced AI may be less like an airplane to be debugged and more like an intelligent agent capable of learning, coordinating, cheating, and adapting.

* 00:00 - Introduction

* 07:53 - Two

* 21:20 - Three

https://open.substack.com/pub/astralcodexten/p/nicholas-decker-in-hell?r=67y1h&utm_campaign=post-expanded-share&utm_medium=web

Get full access to Askwho Casts AI at askwhocastsai.substack.com/subscribe