Artificial Intelligence, Cybersecurity and Cyber Defense

Daniel Ventre’s volume in the Wiley-ISTE cybersecurity series arrived in 2020 at a moment when nearly every major military power was publishing a national AI strategy and nearly every defence ministry was promising algorithmic transformation of its cyber operations. Ventre, a research engineer at France’s CNRS attached to the Centre for Sociological Research on Law and Criminal Justice Institutions and a fixture in the Paris-based cyber-policy ecosystem, treats this moment with the wary patience of a scholar who has watched several previous waves of technology — information warfare, network-centric warfare, “cyber” itself — pass through the doctrine documents of the great powers and leave behind less than promised. The book is a continuation of his long-running project at Wiley-ISTE; earlier titles in the same imprint, including Information Warfare and Cyber Conflict, established the method: read the official texts carefully, compare claims to capabilities, and resist the marketing.

The conversation Ventre joins in 2020 is crowded. The American Department of Defense had stood up the Joint Artificial Intelligence Center in 2018; China’s State Council had published the New Generation AI Development Plan in 2017 setting 2030 as the date for “world-leading” status; the Élysée had launched Cédric Villani’s AI for Humanity strategy in 2018; the European Commission had followed with its White Paper on Artificial Intelligence in early 2020. Every NATO member was scrambling to put numbers and timelines on AI investment. Ventre’s contribution sits among other 2020-era surveys — Forrest Morgan’s RAND volume Military Applications of Artificial Intelligence, Michael Horowitz’s writings on AI and international stability, the run-up to Paul Scharre’s later Four Battlegrounds — but its angle is the cyber domain in particular, and its method is documentary rather than scenario-driven.

The argument the book makes, taken across its chapters, is that the relationship between AI and cybersecurity is symmetrically asymmetrical: AI is being asked to do far more than it can presently do for defenders, while it is already doing useful work for attackers — speeding up reconnaissance, automating spear-phishing, generating plausible content for influence operations. The hype around defensive AI, in Ventre’s reading, is largely the marketing of vendors and the aspirational language of policy documents; the operational reality is closer to assistance and triage than to autonomy. Meanwhile the national strategies, when read carefully, are less coherent than their press releases suggest. They mix industrial policy, military investment, university funding, and ethical posturing in different proportions depending on which ministry holds the pen. Sorting out what is actually being built from what is being declared is, for Ventre, the analyst’s main job.

The book is organised in a way that mirrors that argument. It opens with definitions and a historical chapter on AI itself — symbolic systems, the AI winters, the rise of statistical and connectionist approaches, the deep-learning breakthroughs of the 2010s — so that a non-specialist reader can place the contemporary moment on a timeline that is longer than the news cycle. From there it moves to AI inside cybersecurity proper: how machine learning is used in intrusion detection, in malware classification, in spam and phishing filters, in user-behaviour analytics. Ventre walks through the data dependencies, the false-positive economics, and the adversarial-machine-learning problem — the now-familiar finding that classifiers can be fooled by carefully crafted inputs, which is awkward for anyone proposing to put a classifier at the front of a defensive perimeter.

A central block of the book turns to AI as an offensive instrument. Ventre catalogues the ways automation and machine learning are reshaping the threat landscape: automated vulnerability discovery, large-scale credential testing, generative-text spear-phishing, deepfake audio and video used in social engineering, AI-driven content for influence operations. Here he draws on the literature of the period — the Brundage et al. Malicious Use of Artificial Intelligence report from 2018, work from the Center for a New American Security and from the European Union Agency for Cybersecurity, academic papers on adversarial examples — and on incidents that had become public by the time of writing. The Mirai botnet, the proliferation of off-the-shelf offensive tooling, the DARPA Cyber Grand Challenge of 2016 in which autonomous systems competed at capture-the-flag, all serve as reference points for what automation in offensive cyber actually looks like once the marketing layer is stripped away.

The longest stretch of the book is a country-by-country survey of national positions. The United States is read through the 2018 DoD AI Strategy, the JAIC’s stand-up, the National Security Commission on Artificial Intelligence under Eric Schmidt and Robert Work, and the broader competition framing that came to dominate Washington’s view of China after 2017. Ventre traces the doctrinal language carefully — terms like algorithmic warfare, decision advantage, human-machine teaming — and notes which agencies actually have budgets behind the rhetoric. China is treated at length through the New Generation AI Development Plan, the civil-military fusion framework, and the ecosystem of national champions: SenseTime, iFlytek, Megvii, Cloudwalk, the work of the Academy of Military Science. He pays particular attention to the surveillance applications, since these are the cyber-relevant ones in the Chinese case, and to the export of facial-recognition systems to other authoritarian governments. Russia gets its own treatment built around the Gerasimov-era information-warfare doctrine, the work of the GRU’s information-operations units, and the comparatively modest investment in machine-learning research compared to the United States and China.

The European chapters are where a French researcher’s perspective shows. Ventre reads the Villani report and the subsequent Stratégie nationale pour l’IA closely, looks at the German Federal Government’s AI strategy and the United Kingdom’s Office for Artificial Intelligence, and tracks the European Union’s evolution from the 2018 Coordinated Plan on Artificial Intelligence to the 2020 White Paper. He notes the recurring European pattern: large ambitions, smaller budgets than the Americans or the Chinese, an emphasis on ethics and on trustworthy AI, a worry that the continent will end up as a regulator rather than a builder. The cybersecurity dimension here is the question of digital sovereignty — whether European defenders can use machine-learning tools without depending on American clouds and American models — which Ventre treats as the real strategic question hiding under the ethics language.

Smaller country sections cover Israel, where the relationship between Unit 8200 and the country’s AI start-up ecosystem produces what Ventre describes as a uniquely tight cyber-to-commercial pipeline; India, with the NITI Aayog strategy and the country’s growing engagement with offensive and defensive cyber capability; Japan and South Korea, where the AI investment is heavy but the cyber-defence dimension is muted by post-war doctrinal constraints; and a tour of NATO’s own early work on AI, the discussions that would later mature into the 2021 NATO Artificial Intelligence Strategy.

A separate set of chapters takes up the application areas where AI and cyber intersect with the broader information environment. Influence operations and disinformation get sustained treatment — the 2016 US election, the Brexit referendum, the work of the Internet Research Agency, the Cambridge Analytica episode — read through the lens of what automation actually contributed versus what was done by humans. Ventre is careful here: he is sceptical of the more breathless accounts of AI-powered propaganda and points out that the actual machine-learning content of these operations, at the time of writing, was thinner than the headlines implied. Generative models were still mostly producing curiosities rather than convincing campaign content; the bottleneck was distribution and targeting, both of which had been automated for years on commercial advertising platforms without anyone calling it artificial intelligence.

The book also engages with the ethical and legal architecture being constructed around AI in the cyber and military domains. The EU’s High-Level Expert Group on AI and its Ethics Guidelines for Trustworthy AI, the OECD AI Principles, the work of UNESCO, the Group of Governmental Experts at the United Nations discussing lethal autonomous weapons systems — all are mapped, with attention to the gap between the principles being declared and the practices actually being adopted. Ventre is sympathetic to the project but notes the recurrent pattern in which the states most active in setting ethical norms are also the states with the fewest deployed capabilities to constrain.

Reception of the book within the field has been respectful rather than transformative. It sits in a particular niche — the careful, documentary, French-academic survey — and is read mostly by other scholars working the same beat and by cyber-policy professionals who need a structured overview of national positions as of 2020. Reviewers in the cyber-security literature have praised the breadth of national coverage and the discipline of the reading of primary documents; the criticisms that have surfaced concern the absence of deeper technical case studies of specific machine-learning-on-cyber deployments and a sometimes encyclopaedic quality that prioritises completeness over interpretive sharpness. Where Paul Scharre’s later Four Battlegrounds chose narrative and personal reporting and Kai-Fu Lee’s AI Superpowers chose a single competitive frame, Ventre chose breadth and document-fidelity. That choice is the book’s strength and its limit.

Read today, the book functions as a snapshot of where the conversation stood at the moment the COVID-19 pandemic began. The strategies Ventre dissects have all been updated since: the United States ran the National Security Commission on Artificial Intelligence to its 2021 report, replaced the JAIC with the Chief Digital and Artificial Intelligence Office in 2022, and absorbed the lessons of Ukraine into a fresh round of doctrine; China issued further plans and faced US export controls on advanced chips; the EU moved from the White Paper to the AI Act; NATO published its dedicated AI Strategy. The arrival of large language models from late 2022 onward changed both the offensive and defensive sides of the cyber picture more than anything Ventre’s sources had anticipated. A reader coming to the book in the mid-2020s should therefore treat the country chapters as a baseline rather than a current map, and the technical chapters as an explanation of why the 2020 settlement looked the way it did.

For someone reading widely in this area, the book pairs well with the more technically oriented adversarial-machine-learning literature and with the operationally oriented work coming out of CNAS, the IISS, and the various NATO centres of excellence. It does not replace the specialist literature on cyber operations themselves — Ben Buchanan’s The Hacker and the State or the work of Thomas Rid remain better starting points there — and it does not replace the China-AI specialist literature led by Elsa Kania and Jeffrey Ding. What it offers is a single volume in which the policy documents, the technical capabilities, and the national strategies are laid out next to each other and read against one another by an author with the patience to do the comparison.

What is likely to age well in the book is the method: the discipline of starting from the official text, comparing it to demonstrated capability, and noting the gap. That gap, between declared AI ambition in the cyber domain and actual operational use, is the structural feature of this field that has not gone away. What has aged less well is the specific technical baseline. The world before transformer-based language models reads differently in the cyber-influence chapters than it does now, and any contemporary reader will mentally extend the analysis past the book’s evidence. Ventre’s volume remains a careful map of a moment that, in retrospect, was the last quiet year before the generative-AI surge rewrote the conversation it set out to describe.