Ragi Yaser Burhum Warns AI Agents Multiply Cyber Threats Hundredfold
AI agents are accelerating cyber threats globally, warning that sophisticated, autonomous attacks could multiply by up to 100 times, leaving digital infrastructure “naked.”
Peruvian computer scientist Ragi Yaser Burhum warns that malicious AI, capable of relentless operation without human supervision, is making corporate security a critical and immediate fiscal problem for public and private entities, according to El Comercio Perú.
The swift, automated nature of these attacks marks a fundamental shift in digital risk management. While early security breaches often required human intervention, the emergence of AI-driven agents allows threats to operate continuously and scale attacks by examining multiple systems simultaneously and without rest. This technological leap demands immediate attention from corporate boards and digital security architects.
The Rise of Autonomous Agent Attacks
The threat is not theoretical; it is already being deployed against critical systems. In August 2026, various groups began exploiting networks across the financial, energy, and defense sectors using the tool Artex, which the FBI warned about on November 13, 2026.
Artex allows attackers to convert written objectives into autonomous AI agent commands. These agents execute tasks independently, often achieving privileged access to Microsoft 365 environments of more than 2,250 clients using cloud backup services, according to the FBI. This capability allows malicious actors to move beyond simple manual phishing efforts.
Google’s Threat Intelligence Group noted in a report that cybercriminals are using AI agents to automate every phase of their operations—from identifying vulnerabilities to executing the final exploit.

The speed of these threats is unprecedented. In some cases documented by the FBI, attackers have moved from selecting a target to obtaining administrator access in less than three hours.
A key contributing factor is the availability of modified AI models. While major developers like Anthropic and OpenAI have “guardrails”—safety restrictions that prevent models from providing instructions on how to hack—there exist modified versions known as ‘obliterated’ models that lack these restrictions, according to El Comercio Perú.
These agents don’t work in isolation. According to El Comercio Perú, an AI model handles the reasoning and vulnerability identification, while auxiliary tools, termed ‘harnesses,’ provide the necessary action-execution tools. This combination allows an agent to identify flaws, execute actions, and adjust its strategy without requiring human instructions at every step.
The Technical Debt and Vulnerability Multiplier
The global digital infrastructure faces a massive technical debt, as many foundational systems were built decades ago without modern security protocols. This legacy code presents a fertile ground for AI exploitation.
The primary danger lies in “zero-day” vulnerabilities—security flaws unknown to the system’s developers or those for which no patch exists. AI agents can search for these undetected errors continuously and tirelessly, a capacity far exceeding human capabilities. El Comercio Perú states, “The AI allows it to discover errors that have remained hidden for decades and which can now be exploited at a speed superior to human capability.”

As a cost consideration, the barrier to entry for deploying these attacks is significantly lowering. El Comercio Perú estimates that running a configuration of eight high-performance processing units in the cloud could cost around 68 dollars per hour, totaling approximately 200 dollars for a three-hour operation.
This scalability means threats could multiply dramatically. El Comercio Perú warns, “If you look at hacks of past years, that’s X; by 10, by 100, that is what you’re going to have now.”
The challenge for financial institutions and critical infrastructure hinges on maintaining resilience. Experts suggest that for organizations to effectively counter these threats, they must implement continuous monitoring and robust cyber risk management consulting to proactively scan for hidden flaws.
Developing Sovereign Digital Defense
Facing this escalating risk, the Peruvian expert, Ragi Yaser Burhum, advocates for a strategic, proactive national response. The goal is to develop a sovereign AI model tailored to the nation’s specific security needs, rather than relying solely on off-the-shelf commercial tools.
This sovereign approach would involve adapting open-source ‘obliterated’ models to serve as an internal defensive mechanism. These AI agents would be tasked with pre-emptively scanning and correcting systemic vulnerabilities before they can be leveraged by external attackers. This requires significant investment in AI-driven enterprise security solutions.
The need for specific technical controls is also highlighted by security researchers. For any AI tool accessing corporate repositories, it is essential to classify information, apply confidentiality tags, and implement encryption in transit and at rest, as advised by the Agency Andina, which noted the importance of BYOK (Bring Your Own Keys) policies, according to Agencia Andina.
As the threat landscape evolves, organizations must ensure their security posture is managed by specialized digital transformation and governance services to address the complex interplay between rapid technological adoption and systemic vulnerability.
Burhum, a Peruvian computer scientist from Lima now based in Miami, has spent more than twenty years in the technology industry. He is chief technology officer of Canaria, which builds AI-powered compliance and anti-money-laundering tools for crypto companies, and founder of SeedPhrase Fund.
Earlier in his career he worked in GIS technology at Esri, Microsoft and NVIDIA, and software he wrote at Esri is still used by most countries for census data collection and by U.S. agencies including NASA and the Department of Defense.
On November 19, 2026, Burhum announced he would formally present his findings to the Senate’s Commission on Defense and Internal Order, detailing the risks of agent-based attacks.