On 7 May 2024, The Telegraph reported that Microsoft launched a top secret AI tool for US spies. This marks a significant milestone in AI technology but also the world of espionage. Unlike most AI models which rely on cloud services for learning and inferring patterns from data, this system is entirely isolated from the internet,ensuring a high level of security.
William Chappell, Microsoft’s Chief Technology Officer for Strategic Missions and Technology, described the system as having an “air-gapped” environment. Such is a feature designed to deliver a truly secure platform to the US intelligence community. This development is particularly crucial as the race to integrate generative AI into intelligence data intensifies.
The integration of artificial intelligence into espionage has been a topic of discussion within the intelligence community in recent years. This is because AI has the potential to revolutionize the way intelligence operations are conducted. Particularly by providing enhanced data analysis, threat detection, and decision-making.
Further, the ability to process information piles quickly and accurately can give intelligence agencies a significant edge over adversaries. This then enables them to identify and respond to threats more efficiently. Reaffirming that technology is an important aspect of the intelligence field today, a 2021 CSIS report claimed that “The primary obstacle to intelligence is not technology, it is culture.” (source) (source)
1.0 HUMINT IN A TECH DOMINATED WORLD
In an interview with the Stanford Publication on Human-Centered AI, Amy Zegart states the relevance of human intelligence (HUMINT) gathering in today’s technologically advanced world.
Zegart asserts that when supported by AI algorithms, human operatives can process intelligence more rapidly. And, focus on decision-making tasks that machines are not equipped to handle. “Satellites and AI algorithms are good at counting the number of trucks on a bridge, but they can’t tell you what those trucks mean,” she explains.
This distinction highlights the critical role of human intuition and contextual understanding in intelligence work. A task force dedicated to the integration of AI in espionage concurs. And, that advocates for an approach where AI aids human analysis. They emphasize that human analysts, despite their expertise, are prone to cognitive biases and assumptions. (source)(source)
By incorporating AI into the analytical process, these biases can be mitigated, resulting in more comprehensive and balanced intelligence outputs. AI can provide alternative perspectives, enriching the analytical process and ensuring a more robust evaluation of intelligence data.
However, while AI can significantly enhance the efficiency and depth of intelligence operations, it cannot supplant the foundational culture of the intelligence community. This community thrives on the collection of various forms of intelligence (INTs)—HUMINT, OSINT, SIGINT, GEOINT, MASINT, and IMINT. Each of these disciplines stands to benefit from AI integration. That leaves potential for restructuring traditional methods and more advanced intelligence capabilities.
The Intelligence community however, may be resistant to the changes needed to incorporate AI into these sectors. (source)(source)
1.1 The Fate of Human Spies
Despite the transformative potential of AI in espionage, experts like Richard Moore, director of the UK’s MI6 intelligence agency, and scholars such as Zegart, argue that the human factor in intelligence work remains irreplaceable. AI can efficiently process and analyze vast amounts of open-source data and flagging discrepancies that might otherwise go unnoticed.
However, it still falls to human agents to interpret these findings and convert them into actionable intelligence. For instance, AI might identify unusual activity at a military base. However, being able to understand the significance of this pattern and deciding on subsequent actions require human judgment.
Moreover, AI cannot replicate the value of human relationships MI6 Director Moors states that “… there will always be an extraordinary bond that allows one person genuinely to confide in another, united by a sense of common humanity and purpose”. Human spies play a crucial role in influencing decisions within governments or terrorist groups, leveraging their ability to build trust and extract secrets that AI cannot reach.
As Moore stated, the combination of human skills with AI and bulk data enhances intelligence operations. AI will undoubtedly augment the capabilities of intelligence agencies. Strategic influence that human spies bring to the table are irreplaceable, ensuring their place in the field of espionage.(source)(source)
2.0 OSINT The King of Intelligence in the Tech Age
OSINT refers to open-source intelligence as it gathers and analyzes data which is publicly available. OSINT gathering can be done manually by humans or through machine learning solutions that are able to crunch and sort data through algorithms.
Human led OSINT gathering has proven to be beneficial. And, is accessible to all individuals including ones that do not have security clearances. For instance, In 2012, Siegfried Hecker and Frank Pabian identified the locations of North Korea’s first two nuclear tests. They used commercial imagery and publicly available seismological data. Six years later, their assessments were confirmed when the North Korean government disclosed the test sites.
Similarly in July 2020, researchers David Albright and Fabian Hinz, using geolocation tools and commercial satellite imagery identified a fire in Iran as an explosion in the nuclear facility. The fire was bright enough to be detected by a weather satellite and the Iranian Atomic Energy Organization stated it was a minor incident.
Albright and Hirz however, working in two different NGOs and conducting individual research revealed the fire was near a nuclear centrifuge assembly building in Natanz. Their findings, disseminated on Twitter and were quickly picked up by major news outlets. These are examples as to how OSINT has been valuable in recent years and display the accessibility of such information. (source),(source)
2.1 Re-conceptualizing OSINT
A CSIS report released in 2021, provides suggestions to the US intelligence community to “reconceptualize OSINT” as a crucial factor for US intelligence gathering. In the ever-evolving landscape of intelligence gathering, OSINT has emerged as a resource, revolutionizing the capabilities and of strategic intelligence collection.
The growing quality, relevance, and timeliness of OSINT are now fundamental to all-source analysis, positioning it as a foundational component on par with the traditional INTS.The integration of cloud computing, AI, and advanced analytics has amplified the power of OSINT. It provides high-quality, multisource intelligence analysis to be produced at the unclassified level from virtually anywhere. The CSIS report from 2021, suggests a the formation of an OSINT agency/ entity which could serve as the public face of the intelligence community.
This agency would use AI and Machine learning models to crunch, sort, classify, and identify intelligence and communicate it with the other agencies to help build a robust intelligence community. (source),(source)
3.0 Collaboration among Nations
On 15 November 2023, “Five AIs Act” was introduced by Representatives Mike Gallagher and Ro Khanna. This act highlights the evolving landscape of international collaboration in the realm of artificial intelligence (AI). This proposed legislation aims to strengthen the Five Eyes alliance through creating a dedicated working group to drive AI innovation and interoperability. The “Five AIs Act” seeks to further this collaboration by establishing a “Five AIs Strategic Artificial Intelligence Working Group.”
This group would be tasked with accelerating AI experimentation, governance, and deployment efforts across the alliance. The bill mandates that the U.S. Secretary of Defense, in coordination with the Director of National Intelligence, launch this group, designating a senior civilian or military official to lead it. The bill also encourages participation from members of the Five Eyes Intelligence Oversight and Review Council.(source)
The working group’s responsibilities encompass identifying potential solutions to improve the interoperability of AI systems used in intelligence sharing, battle-space awareness, and other operational contexts.
Additionally, the group would formulating a unified strategy for the five nations to collectively research and deploy AI technologies. This strategy aims to capitalize on commercial products to strengthen immediate military cooperation. This initiative highlights the importance of international cooperation in addressing global AI-driven threats.(source)
4.0 Cyber Espionage and AI
The integration of generative AI into cyber operations has paved the way for sophisticated forms of cyber espionage. As highlighted by Microsoft and cybersecurity experts,countries such as North Korea, Iran, Russia, and China are increasingly leveraging large-language models (LLMs) like OpenAI’s ChatGPT to enhance their offensive capabilities.
According to Microsoft, these countries have employed AI to conduct extensive research on targeted entities and to craft highly convincing spear-phishing campaigns. For instance, North Korea’s Kimsuky group has utilized AI to gather intelligence on foreign think tanks. Iran’s Revolutionary Guard has also deployed AI in social engineering and in creating deceptive emails aimed at prominent figures.
Similarly, Russian and Chinese cyber groups have also explored AI’s potential in enhancing technical operations, including researching advanced military technologies and evading network detection measures. Amit Yoran, CEO of Tenable, states “Of course bad actors are using large-language models – that decision was made when Pandora’s Box was opened.”
This sentiment reflects on the fact that the rapid deployment of AI without security considerations has opened avenues for exploitation.(source)
4.1 The China Case
Although the Chinese embassy in the US states that recent reports are “groundless smears and accusations agaisnt China”, Microsoft claims that Chinese hackers have been using AI tools.
According to Reuters, Chinese state-backed hackers were using AI to ask questions about other intelligence agencies, and important individuals within those agencies. Chinese AI powered cyber operations also target critical infrastructure, government agencies, and private enterprises, exploiting network vulnerabilities and employing AI algorithms to analyze stolen data for strategic advantage.
The groups identified by Microsoft Threat Intelligence, including Gingham Typhoon and Flax Typhoon, illustrate China’s strategic objectives. These groups use use AI to enhance their phishing attacks and manipulate public opinion through disinformation campaigns.In addition to gathering intelligence the groups also sow discord and influence on online platforms.
On the other hand, AI and Machine Learning technologies are also used by the U.S. cybersecurity defenses against cyber activities originating from China. Machine learning and big data analytics enable the NSA to identify potential threats hidden under legitimate network activities.
Despite concerns about the dual-use nature of AI in offensive cyber operations, director of the NSA Cybersecuirty Directorate Rob Joyce highlights its significant defensive capabilities. He indicates its effectiveness in safeguarding critical U.S. networks against strategic threats aimed at causing “societal disruption and panic”.(source)(source)(source)
5.0 Conclusion
In the evolving landscape of espionage, integration of AI marks an advancement both in enhancing intelligence capabilities and uncovering various challenges. AI-driven technologies have the capability to revolutionize how intelligence agencies collect, analyze, and utilize data.
They also allow for speed and accuracy in processing vast amounts of information. This transformative potential is evident in the deployment of AI for tasks ranging OSINT gathering to the use of large-language models conduct targeted espionage. However, although these technological developments allow for efficiency, the human element is indispensable.
Moreover, international collaboration is also important in mitigating the risks posed by AI in espionage. By aligning strategies for AI innovation and governance, nations can collectively strengthen defenses against emerging common threats.
5.1 The dangers of AI in Espionage
While AI amplifies the potential for intelligence gathering and analysis, its integration must be guided by ethical considerations and security measures.AI’s reliance on machine learning algorithms introduces vulnerabilities that malicious actors can exploit to evade detection.
This allows actors to be able to manipulate AI systems, leading to potential breaches of classified data.The use of generative AI in espionage also introduces significant threats as it can fabricate convincing fake audio, video, and textual content.
It then makes it challenging to identify true intelligence from generated(false) intelligence. This enables malicious actors to manipulate public opinion and destabilize societies by disseminating deceptive narratives that exploit existing vulnerabilities. Therefore, the dangers that come with integrating AI in espionage also warrant robust safeguards and international norms to mitigate risks.