— Lawmakers from both parties press the Pentagon for answers on the Iran war: strategic objectives, the scope of the commitment, and the absence of formal congressional authorisation.
— Research from RAND and the Nuclear Threat Initiative warns that AI can bridge the informational gaps that once made biological attacks fail — turning a speculative risk into a practical one.
— Internal documents show a Chinese hacker-for-hire firm, ZRON, selling stolen foreign intelligence to state security clients and using AI to make it actionable: espionage as an automated service.
I.
Congress demands answers as Pentagon defends the Iran war
Lawmakers from both parties have expressed frustration over the Pentagon's lack of transparency regarding the escalating military campaign against Iran. During recent House and Senate committee hearings, Defense Undersecretary Elbridge Colby faced sharp questioning about the administration's strategic objectives, the scope of the commitment, and the absence of formal congressional authorization for military operations.
Colby has defended the campaign as a pragmatic, short-term operation aimed at degrading Iranian capabilities. "This is not another Iraq War, this is not nation-building," he told the House Armed Services Committee, describing the approach as that of "flexible realists." Yet Democratic lawmakers pointed to historical parallels, with U.S. Representative Sara Jacobs (D-Calif.) noting that President George W. Bush had used nearly identical language at the onset of the Iraq War in 2003 — a conflict that lasted eight years and cost thousands of lives and trillions in taxpayer dollars.
The Pentagon estimates that military operations to date have cost $37.5 billion, and the Trump administration is now requesting additional appropriations. Meanwhile, some Republican senators have raised concerns about escalation and the lack of a defined exit strategy. Rep. Sarah Elfreth (D-Md.) wrote that "we are once again attacking Iran without more clarity, legal justification, or consultation of Congress." Colby acknowledged a prior statement in which he wrote that "the threat from Iran pales in comparison to that from China," but asserted that he remains supportive of scoped military objectives. As November's midterm elections approach, the debate has intensified, with Democrats linking military spending to rising costs of living and Republicans questioning the Pentagon's candour with Congress.
Artificial intelligence and the bioweapons convergence: a new vulnerability
Research published by the RAND Corporation and analyses from the Nuclear Threat Initiative have raised alarm about the convergence of artificial intelligence and biotechnology in weapons development. The concern is not speculative. A previous biological attack that failed because of informational gaps might succeed in a world in which AI tools can bridge that gap by aggregating and synthesizing open-source information on pathogen engineering, viral replication, and weaponizable bioagents.
Elizabeth Sherwood-Randall, who served as the White House's Homeland Security Advisor and Deputy National Security Advisor from 2021 to 2025, has been among the most vocal advocates for understanding this threat. "Without robust safety measures," the research warns, "AI could enable bad actors to perform dangerous research, such as the development of bioweapons." The advantage to an attacker is asymmetric: the information barriers that once protected biological security have collapsed. An AI model trained on scientific literature, academic publications, and publicly available biological research can now provide guidance that would previously have required specialized expertise, years of laboratory work, or access to classified research.
The policy response has been fragmented. Experts have recommended the establishment of an international "AI-Bio Forum" to develop model guardrails, new governance structures that move faster than traditional arms-control regimes, and strengthened biosecurity controls at the interface between digital design tools and physical laboratory systems. The timeline is compressed: the authors of these reports emphasize that the window for preventive policy is narrow, and that AI capabilities are advancing faster than regulatory oversight. An information gap that protected biological security for decades can close in months.
From hacking-for-hire to automated intelligence: China's AI-powered espionage
Internal documents obtained by The Wall Street Journal reveal that a Chinese hacker-for-hire company, Zhengzhou Zhirong Network Technology Co. (ZRON), is selling stolen foreign intelligence directly to Chinese government security clients, and using artificial intelligence systems to make the data actionable for police and intelligence agencies. The operation represents a qualitative shift in Chinese cyber espionage: what was once purely manual intrusion work is now being automated and scaled with AI models.
The company offers menu-based intelligence derived from compromised government email systems of China's rivals and ostensible allies alike. ZRON has deployed AI models to parse, index, and synthesize stolen data into formats useful for operational intelligence. The advantage for Chinese security agencies is immediate: a human analyst would require years to organize and analyse billions of pieces of data. An AI system can reduce that task to hours. China's cyber theft programme has been so successful and vast that it has created an informational abundance problem — too much raw data for human handlers to process. AI solves that problem by converting stolen information into intelligence.
The scale of the programme extends beyond ZRON. Chinese government hackers have been linked to breaches affecting hundreds of millions of records from firms including Marriott International, Equifax, and Anthem. More recently, state-sponsored actors have been observed using open-source AI tools to compress intrusion timelines from weeks to under six hours. The FBI has noted that Chinese entities account for more than half of all state-sponsored cyberattacks against American technology companies, with particular focus on theft of AI capabilities themselves — a form of espionage aimed at narrowing the technological gap with the United States. The convergence of massive data theft and AI-driven analysis has created a new form of national security vulnerability: not the breach itself, but the ability to extract meaning from it faster than defenders can respond.
Key terms — et leur équivalent français
hacker-for-hire
pirate informatique professionnel, entreprise de hacking
to compromise (a network)
compromettre, s'introduire et modifier (un réseau)
Conventional security models treat physical or technical barriers as the primary defense — firewalls, encryption, access controls. The convergence of three technologies — abundant data theft, artificial intelligence, and open-source research — is collapsing a different kind of barrier: the information barrier, the gap between raw data and actionable intelligence. When that barrier collapses, stolen information becomes intelligence, and the attacker no longer needs an expert; the AI does the expert work. This shift changes both cyber defense and biodefense, and demands rethinking how governments classify and protect not just secrets, but the structural gaps that have historically defended against large-scale harm.
A new issue every Monday
The week's essentials in defence, security and intelligence, assessed from open sources: in English, with key terms translated into French.