# France's ANSSI Publishes Seven Security Principles for Enterprise AI Agents > ANSSI published a two-page guide with seven security principles for enterprise AI agents, covering risk assessment. - Published: 2026-10-31T19:25:43+00:00 - Section: Regulation & Approvals - Author: News Desk - URL: https://newsforagent.com/news/2026-10-31/anssi-seven-security-principles-ai-agents - Source: [ERP News](https://erpnews.com/frances-anssi-sets-out-seven-security-principles-for-enterprise-ai-agents/) ## Key takeaways - ANSSI published an English-language guide, Securing an Agentic AI System, on October 30 as part of its Back to Basics collection. - The two-page guide sets out seven principles, including controls agents cannot override and human approval for critical actions. - ANSSI recommends checking whether a sequence of individually authorized actions could produce a prohibited outcome. France's national cybersecurity agency, [ANSSI](https://akta.pro/company-profiles/000cgxs/anssi), published seven security principles for organizations deploying AI agents on October 30, recommending controls agents cannot override, human approval for critical actions and mechanisms to stop abnormal activity. [ERP News](https://erpnews.com/frances-anssi-sets-out-seven-security-principles-for-enterprise-ai-agents/) reported the publication. ANSSI is France's national cybersecurity authority, a government agency under the Prime Minister that provides cyber defense, incident response through CERT-FR, certification through NCC-FR, training through CFSSI and regulatory oversight for state entities and critical infrastructure operators. It was founded in 2009 and is headquartered in Paris. ## What the ANSSI Agentic AI Guide Covers The English-language guide, Securing an Agentic AI System, addresses organizations deploying, configuring or integrating agentic systems, including commercial products whose settings and connectors they control. It is part of ANSSI's Back to Basics collection, which primarily serves cybersecurity professionals such as security leaders and architects. The publication presents technical recommendations rather than a new regulatory obligation or product certification. The two-page guide organizes its recommendations around seven areas: assessing risks throughout the system, starting during design and updating after changes; limiting autonomy by documenting the mission and expanding capabilities gradually; classifying actions into autonomous execution, mandatory human approval and prohibited activity; isolating development, evaluation and production identities, connectors and memory; enforcing authorization independently by checking permissions and business rules outside the model and denying actions by default; containing operational impact by protecting secrets, limiting agent interactions and providing emergency shutdown and recovery mechanisms; and maintaining demonstrable control through testing, monitoring, component inventories and protection of activity logs. ANSSI also recommends checking whether a sequence of individually authorized actions could produce a prohibited outcome. Human approvers should see the exact proposed action, its purpose and expected effects. In a hypothetical procurement workflow, creating a supplier record, changing payment details and preparing a payment could each be permitted activities, and their combination deserves separate scrutiny, particularly where the organization relies on separation of duties. ANSSI recommends preventing recovered systems from repeating actions that have already produced an external effect, such as checking transaction status before retrying an interrupted payment task. ## ANSSI's Earlier Guidance and Policy Moves According to akta.pro [company records](https://akta.pro/company-data), ANSSI published a crisis exercise kit for the agri-food sector in June 2026, extending crisis preparedness resources into a new vertical outside traditional defense and digital infrastructure sectors. The same records show ANSSI published a training framework dedicated to remediation management piloting that month, formalizing guidance for post-incident remediation governance. ANSSI also issued a call for expression of interest on the security of cybersecurity ecosystems in June 2026, signaling an industrial policy push to strengthen the domestic cybersecurity supply chain. The records describe ANSSI as the national authority for cybersecurity and cyber defense across all of France, with a nationwide operational mandate for protecting digital systems of national interest. ## ANSSI's Recent Incident Response Work On September 30, 2026, 01net.com reported that ANSSI released its report on the DGFIP cyberattack, finding it exploited known weaknesses in identity, architecture and detection; the attacker stole dozens of agent credentials and exfiltrated 14 gigabytes of data over two waves. On October 1, 2026, 01net.com reported that ANSSI's innovation lab and the state's digital ministry were hit by a cyberattack exploiting a Metabase vulnerability, compromising 118 accounts; the breach occurred on August 6, 2026, exposed administrative data, and ANSSI has since patched the instances and reset passwords. On October 5, 2026, 01net.com reported that ANSSI was hacked via a Metabase software flaw and the Paris prosecutor's office opened an investigation examining data access and extraction. ## What Comes Next for ANSSI On June 16, 2026, ANSSI confirmed it will cease certifying security products lacking quantum-resistant encryption, according to Quantum Zeitgeist, establishing a deadline for post-quantum cryptography adoption. The agency's [English-language site](http://www.ssi.gouv.fr/en) carries its published guidance. ## Companies in this story - [ANSSI company profile](https://akta.pro/company-profiles/000cgxs/anssi) · website: http://www.ssi.gouv.fr/en --- News for Agent · https://newsforagent.com/llms.txt