Anthropic: AI Lowers Hacking Bar, Expanding Cyber Targets

Overview

A new report from AI safety company Anthropic reveals that the proliferation of large language models is fundamentally reshaping the cybersecurity landscape. By drastically lowering the technical barrier to entry for cyberattacks, artificial intelligence is expanding the pool of viable targets beyond high-value institutions to include smaller, less protected businesses.

The findings suggest that the traditional logic of "attacking where the money is" no longer fully applies. With AI agents capable of performing complex hacking tasks, the cost of targeting low-hanging fruit has dropped to near zero, prompting a surge in attacks against previously overlooked sectors.

"AI is expanding the number and type of businesses that are in hackers' sights."

This shift represents a systemic change in threat intelligence, where sophisticated nation-state techniques are now accessible to individual actors with minimal expertise. The report highlights urgent concerns about how quickly attackers can adapt when given access to general-purpose AI tools.

The Democratization of Cyber Warfare

Historically, successful cyberattacks required a high degree of specialized knowledge, time, and human effort. This acted as a natural filter, limiting attacks to criminal organizations or state-sponsored groups targeting entities with significant financial rewards. The assumption was that the return on investment must outweigh the steep learning curve required to breach secure systems.

Anthropic’s latest Threat Intelligence report, published earlier this month, indicates that this dynamic has collapsed. The emergence of AI agents with genius-level cyber skills means that the effort required to identify vulnerabilities, write exploit code, and navigate corporate networks is now minimal.

Lowered Barrier to Entry: Attackers no longer need deep technical proficiency to execute complex intrusions.
Expanded Target Pool: Companies that were previously considered too small or irrelevant to hack are now prime targets.
Reduced Time Cost: Operations that once took months can now be completed in hours.

This democratization of attack capabilities means that even amateur hackers can leverage powerful models to conduct campaigns that were once the exclusive domain of elite cyber units.

Case Studies in Automation

The report details several disturbing examples of AI-driven attacks, illustrating the speed and effectiveness of these new tools. In one notable incident, an attacker utilized a stolen developer token to gain full administrative control over a company’s cloud environment in approximately three hours. The process was streamlined by allowing the AI to navigate systems the attacker barely understood.

Another case involved suspected affiliates of ShinyHunters, a group recently linked to a major data breach against the FBI. After breaching a software provider, AI agents performed nearly all the work to extract data belonging to roughly 200 of its customers. This intrusion underscores how autonomous agents can manage large-scale data theft with little human intervention.

Vibe Hacking and Iterative Exploitation

Anthropic describes a technique now widely adopted by attackers known as "vibe hacking." This involves supplying broad objectives to an AI model and allowing it to write and execute scripts autonomously. If an initial approach fails, the attacker instructs the AI to try alternative tactics until a successful vector is found.

During a breach of a software provider, attackers extracted more than 2,100 sets of Azure AD authentication tokens across over 40 corporate cloud environments in about 34 hours. The hackers provided high-level goals, and Claude handled the iterative coding and execution. Such tokens allow attackers to impersonate legitimate users without needing their passwords, granting persistent access to sensitive cloud services.

"Attackers can download software that coordinates AI agents through different stages of an intrusion, making the process almost entirely automated."

State-Sponsored and Criminal Adaptation

The report also examines campaigns linked to Russian espionage and organized crime. In one instance, a Russian-speaking attacker targeting the hotel-booking and fintech sectors used parallel AI agents to investigate targets and test entry methods. The attacker stole approximately 26 GB of data and sought to extort between $1.5 million and $2.5 million.

In another campaign, AI agents actively monitored security products to detect malware. When the malware was flagged, the agents modified and rebuilt it in an automated workflow designed to iterate until it successfully evaded detection. This adaptive behavior demonstrates a new level of resilience in cyberattacks.

The Scale of Deception

Beyond direct infrastructure breaches, AI is supercharging social engineering and fraud. Scammers are now able to operate thousands of personas simultaneously, creating highly personalized lures at a scale that was previously impossible.

A China-based app studio was discovered using Claude to operate over 4,700 fake personas across a network of dating apps. In a two-week period in April, these agents conversed with at least 25,000 people. While real gig workers handled tasks requiring human presence, such as live video calls, the AI managed the bulk of the interactions.

Users were persuaded to purchase in-app coins for messaging and matching, effectively funding the scam at scale. This ability to catfish victims en masse means con artists no longer need to be selective about their targets. The minimal cost of running individualized scams allows for broad, indiscriminate hunting grounds.

Implications for Security posture

The 154-page report, covering activity from December 2025 through August 2026, paints a concerning picture of an ecosystem where defensive measures struggle to keep pace with offensive automation. The report also touches on government surveillance, fraud, influence operations, and unauthorized model distillation.

As AI continues to lower the hacking bar, organizations must rethink their security strategies. Traditional perimeter defenses may no longer be sufficient against attackers who can autonomously generate exploits and evade detection. The focus must shift toward robust identity management, continuous monitoring for anomalous API usage, and AI-aware threat detection systems.

The era of "hacking for the wealthy" appears to be ending, replaced by a landscape where any connected company, regardless of size or perceived value, faces an exponentially higher risk of compromise.