# Risk Reduction Framework_Sept 2026 **By:** Jack Shanahan **Published:** 2026-09-19T21:01:04.105Z **Source:** [LinkedIn](https://www.linkedin.com/posts/jackntshanahan_thanks-to-eric-richardson-inhr-team-for-their-share-7507181993370206208-0Ksx) --- # Jack Shanahan’s Post [Jack Shanahan](https://www.linkedin.com/in/jackntshanahan?trk=public_post_feed-actor-name) Thanks to [Eric Richardson](https://ch.linkedin.com/in/eric-richardson-614846ab?trk=public_post-text) [INHR](https://www.linkedin.com/company/inhrgeneva?trk=public_post-text) team for their work in producing this excellent risk reduction framework for agentic AI in military systems. I'm proud to be a part of the team that has been working on this framework over the past six months, along with colleagues from many other countries. It reflects inputs from developers, operators, policymakers, and technical experts. While true agentic AI remains in a nascent stage in all militaries, including within the US, it is only a matter of time before the pressure builds to accelerate adoption. This framework lays out the principles every country needs to take into account when integrating advanced AI agents into military systems, from decision support systems to autonomous and agentic weapon systems. This will have to be a living document, of course, given how fast the technology is changing. But it offers a crucial reference point at a crucial time. [INHR](https://www.linkedin.com/company/inhrgeneva?trk=public_post_reshare_feed-actor-name) INHR is pleased to announce the launch of the September 2026 working draft of the Risk Reduction Framework for Agentic AI in the Military. The Framework was presented last week during the UN Group of Governmental Experts on Lethal Autonomous Weapons Systems (LAWS GGE) in Geneva and is now available on the INHR website. Developed through an international dialogue involving experts from government, academia, industry, and the military, the Framework outlines practical principles for reducing risks associated with increasingly autonomous AI systems in military applications. This September draft builds on discussions held throughout 2026, including the International Cooperation in AI-Enabled Weapons & Security Trilateral Plenary Meeting in Shanghai. The Framework remains a working draft and will continue to evolve through further research, dialogue, and engagement. Read the Framework: [https://ow.ly/E3MI50ZNeCj](https://ow.ly/E3MI50ZNeCj?trk=public_post_reshare-text) [#AgenticAI](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fagenticai&trk=public_post_reshare-text) [#AIGovernance](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Faigovernance&trk=public_post_reshare-text) [#NationalSecurity](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fnationalsecurity&trk=public_post_reshare-text) [#LAWS](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Flaws&trk=public_post_reshare-text) [ Risk Reduction Framework\_Sept 2026 docs.google.com ](https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fdocs%2Egoogle%2Ecom%2Fdocument%2Fd%2F1pe4bX05Kb_2poPiAF_-h5aASl9Vod_wWNrD_zpZLpW4%2Fedit%3Ftab%3Dt%2E0%26usp%3Dembed_facebook&urlhash=5MS6&trk=public_post_reshare_feed-article-content) Risk reduction frameworks for agentic systems must cover both model-level and operational risks. Does this framework address the human-in-the-loop threshold for high-stakes decisions? [Phaedra Boinodiris](https://www.linkedin.com/in/phaedra?trk=public_post_comment-text) - I’m incredibly impressed by the work that this team has put into this framework, and grateful to have had the opportunity to contribute! Especially to those at [INHR](https://www.linkedin.com/company/inhrgeneva?trk=public_post-text) who have supported this work and are asking all the right questions. Militaries are still in the early stages of the pivot toward agentic AI, making this a critical moment to begin serious conversations about the risks and how they can be mitigated before adoption accelerates. I’m very pleased to see this work published and look forward to the conversations that will surely follow. INHR is pleased to announce the launch of the September 2026 working draft of the Risk Reduction Framework for Agentic AI in the Military. The Framework was presented last week during the UN Group of Governmental Experts on Lethal Autonomous Weapons Systems (LAWS GGE) in Geneva and is now available on the INHR website. Developed through an international dialogue involving experts from government, academia, industry, and the military, the Framework outlines practical principles for reducing risks associated with increasingly autonomous AI systems in military applications. This September draft builds on discussions held throughout 2026, including the International Cooperation in AI-Enabled Weapons & Security Trilateral Plenary Meeting in Shanghai. The Framework remains a working draft and will continue to evolve through further research, dialogue, and engagement. Read the Framework: [https://ow.ly/E3MI50ZNeCj](https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fow%2Ely%2FE3MI50ZNeCj&urlhash=s1d6&trk=public_post_reshare-text) [#AgenticAI](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fagenticai&trk=public_post_reshare-text) [#AIGovernance](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Faigovernance&trk=public_post_reshare-text) [#NationalSecurity](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fnationalsecurity&trk=public_post_reshare-text) [#LAWS](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Flaws&trk=public_post_reshare-text) [ Risk Reduction Framework\_Sept 2026 docs.google.com ](https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Fdocs%2Egoogle%2Ecom%2Fdocument%2Fd%2F1pe4bX05Kb_2poPiAF_-h5aASl9Vod_wWNrD_zpZLpW4%2Fedit%3Ftab%3Dt%2E0%26usp%3Dembed_facebook&urlhash=5MS6&trk=public_post_reshare_feed-article-content) - Dangers of AI!! A flawed AI-based intelligence report almost led US forces to intercept a Chinese vessel, highlighting serious risks of artificial intelligence in military decision-making. - [Mark Jennings-Bates](https://ca.linkedin.com/in/jenningsbates?trk=public_post_feed-actor-name) The military solved the correctness problem before AI existed. They called it discipline. Every element of military doctrine is a correctness control. Rules of engagement define exactly what a soldier is authorised to do, under which conditions, against which targets. That is an authority boundary. The chain of command specifies who is responsible for every decision. That is liability assignment. After-action review requires a traceable record of what happened, what authority existed, and whether the outcome matched intent. That is output verification. Doctrine codifies what has been learned from failure. That is governance effectiveness. Officer education ensures leadership understands the difference between what a unit can do and what it is authorised to do. That is correctness awareness. The military did not build this infrastructure because they were cautious. They built it because they learned, through centuries of consequence, that capability without discipline does not produce military outcomes. It produces chaos with weapons. Now we are building AI systems with extraordinary capability and deploying them into defence workflows. The question is not whether AI governance is possible. The question is whether the people deploying AI have read the doctrine that already answers it. Capability transfers. Discipline does not transfer automatically. [#MilitaryAI](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fmilitaryai&trk=public_post-text) [#AutonomousSystems](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fautonomoussystems&trk=public_post-text) [#AIGovernance](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Faigovernance&trk=public_post-text) [#DefenceInnovation](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fdefenceinnovation&trk=public_post-text) [#TheCorrectnessProblem](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fthecorrectnessproblem&trk=public_post-text) [#CommandAuthority](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fcommandauthority&trk=public_post-text) [#MosaicDM](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fmosaicdm&trk=public_post-text) [#NATOInnovation](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fnatoinnovation&trk=public_post-text) [#AIAccountability](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Faiaccountability&trk=public_post-text) [#MilitaryDoctrine](https://www.linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fmilitarydoctrine&trk=public_post-text) - ◈ CURATED NEWS | AI & Superintelligence (SI) After the Trump–Xi summit, China and the United States say they will establish a channel for handling AI-related incidents — and accelerate military crisis communications. A hotline is not a safety standard. But after rogue-agent weeks and UNSC lab briefings, incident notification between the two AI superpowers is no longer optional infrastructure. Source: The Washington Post [https://lnkd.in/eDmrQeZ5](https://www.linkedin.com/redir/redirect?url=https%3A%2F%2Flnkd%2Ein%2FeDmrQeZ5&urlhash=ZfBn&trk=public_post-text) - [Military AI](https://www.linkedin.com/company/militaryai?trk=public_post_feed-actor-name) Canada builds its own AI bench 🛠️⚡ Canada is investing 33.5 million Canadian dollars ($23.9 million) to develop military AI using domestic technology. The funding backs MAIple Stack, which will test tools from Canadian AI and defense firms for possible military use.