Two weeks ago, security technologist and author Bruce Schneier attended the [un]prompted AI security conference held in San Francisco, an event that brought together researchers, developers, and security professionals to examine the rapidly evolving landscape of artificial intelligence. Described as an excellent and intensive gathering, the conference featured three days of dense, highly technical talks covering virtually every aspect of the intersection between artificial intelligence and cybersecurity. Schneier, who delivered a presentation at the event focused on the concept of integrous AI—a topic he indicated will be shared publicly once the recording goes online—highlighted the event as a vital meeting ground for the security community.
Reflecting on the broad discussions that took place during the conference, Schneier noted that the sprawling field of artificial intelligence security can be effectively categorized into three distinct pillars. According to his analysis, it is crucial for researchers to clearly understand which of these specific areas their work falls under, while also recognizing that the broader technical community must dedicate sustained effort to all three simultaneously. These foundational pillars—security of AI, security with AI, and security from AI—served as an underlying framework for the conference, which featured substantive presentations and debates addressing each category in depth.
The first pillar, the security of AI, encompasses the traditional and emerging challenges of protecting artificial intelligence models, training pipelines, and underlying infrastructure from malicious compromise. As organizations increasingly integrate machine learning systems into core operations, these models themselves become high-value targets for adversaries. Security in this domain involves defending against threats such as data poisoning during the training phase, model extraction attacks that attempt to steal proprietary architectures, adversarial inputs designed to trick classifiers, and vulnerabilities within the software libraries and hardware environments that power large-scale machine learning workloads. Ensuring that AI systems operate reliably and resiliently in the face of direct attacks remains a paramount concern for engineers and security architects alike.
The second pillar, security with AI, shifts the perspective to how artificial intelligence can be harnessed as a defensive tool by security practitioners. In this context, machine learning and automated reasoning systems are deployed to enhance human capabilities, ranging from automated threat detection and malware analysis to vulnerability discovery and incident response triage. As networks grow exponentially more complex and the volume of security telemetry overwhelms human analysts, leveraging AI to sift through noise, identify anomalous patterns, and accelerate defensive actions has become an area of intense development. However, this approach also introduces its own set of complications, including the need to ensure that the defensive tools themselves are robust, transparent, and not easily bypassed by sophisticated attackers who understand how machine learning models generalize.
The third pillar, security from AI, addresses the broader societal, geopolitical, and systemic risks posed by the proliferation and misuse of powerful artificial intelligence capabilities. This domain confronts the ways in which bad actors can leverage AI technologies to scale malicious operations—such as generating highly convincing social engineering campaigns, automating the discovery of zero-day software vulnerabilities, or producing sophisticated misinformation at an unprecedented scale. Mitigating security from AI requires policymakers, researchers, and technologists to anticipate misuse vectors, establish governance frameworks, and develop technological safeguards that prevent general-purpose AI systems from being weaponized against critical infrastructure, democratic institutions, or public safety.
The [un]prompted conference provided a timely venue for exploring these intersecting challenges, offering attendees a comprehensive view of how the security industry is attempting to keep pace with the swift advancement of machine learning technologies. By breaking down the discipline into the security of AI, security with AI, and security from AI, industry leaders are better equipped to navigate the complexities of the field, allocate research resources effectively, and collaborate on comprehensive defenses. As the technology continues to mature and integrate deeper into the fabric of daily digital operations, ongoing dialogues at specialized gatherings like [un]prompted will remain essential for shaping a secure and resilient future for artificial intelligence.
Cyber Security & Cryptography
Cyber Security & Cryptography
Cyber Security & Cryptography
