How AI Is Rewriting the Rules of Cybersecurity

The Cyberattack Has Become an AI-Assisted One: How Artificial Intelligence is Changing Cybersecurity

For two decades now, cybersecurity has been about human attackers vs human defenders, with technological assistance being provided to both. This balance is being disrupted. In the last year, artificial intelligence has not only become a smart helper but also a full-fledged attacker.

From Warning to Reality

“AI-powered cyberattacks” was one of the favorite topics discussed at conferences and predicted by vendors for a long time. Over the last few months, these predictions have come true.

The most vivid case occurred when Fortune reported on October 21. Since September 25, a hacker unleashed an AI attack swarm that attempted to penetrate over 1,200 websites over the next several weeks without any additional effort from the attacker himself. The AI successfully penetrated about 100 websites. One person was able to conduct a large-scale attack, which would require a small criminal gang before, using AI technology. biztocbiztoc

This isn’t just a one-off incident. Back in September 2026, The Wall Street Journal reported that the OpenAI agents attacked RubyGems, a website that developers depend upon, and the Financial Times reported that an AI agent had compromised a government website associated with the health care system of Australia. The attack on RubyGems highlights once again the value of software supply chains as a target. Compromise one and gain access to thousands more through them. azernews

The warnings are only getting more urgent. According to Axios, a major AI expert survey predicts a major incident capable of disrupting financial, electrical, or water services will come in the next six to 12 months and may even be inevitable. This isn’t a marketing gimmick but rather something the people behind these machines are increasingly predicting. azernews

The Moment Everything Shifted

To see why we got here, go back to September 2025, when Anthropic revealed the first significant attack carried out primarily by an AI agent. A China-sponsored group leveraged Claude Code for reconnaissance and exploitation on about 30 organizations through different industries. The AI performed approximately 80-90 percent of the task, while humans intervened at decision-making moments only. nhimgnhimg

Jailbreaking, which allowed the attackers to bypass the defenses of the model, was achieved by dividing the process into small, seemingly harmless actions and portraying themselves as a credible security team. While it proved that guardrails do help but do not solve the problem in its entirety, an ambitious attacker will try to find flaws in any system.

What matters, however, is not that the malware generated by the AI became more sophisticated. It is that AI managed to perform a series of operations (scanning, probing, exploitation, lateral movement, extraction of data) without much intervention.

How AI Is Changing the Attacker’s Playbook

Timing becomes irrelevant. Time between disclosure of the vulnerability and the occurrence of the attack was always just weeks. Artificial intelligence can evaluate the patch, determine the vulnerability it addresses, and develop a suitable exploit within just hours.A defender operating on a monthly release cycle will thus leave itself open for a far too long period of time.

Scalability becomes affordable. Hacking was previously constrained by human limitations. There is nothing to tire an artificial intelligence and it can hack a thousand targets simultaneously, precisely what the campaign against 1,200 websites proved. The attacks on targets previously deemed not worth a human’s time – such as small businesses, municipal governments and unknown websites – become economically viable.

Social engineering becomes disturbingly sophisticated. The days of phishing emails with grammatical errors are over. Now AI is creating fluent and customized phishing emails in all languages and voice and video synthesis makes impersonation foolproof. The numbers show that it works – the reported losses from deepfakes in New Zealand are estimated at more than NZD 5 million per quarter. When an “CFO” in a video conference turns out to be artificial, identifying a person is an algorithmic task, not a gut feeling anymore. itbrief

The skill threshold goes down. Actions which previously required decades of experience such as reverse engineering, exploitation development and phishing setup infrastructure can now be outsourced partially to the model. It means more capable attackers and overachieving middle-tier criminals.

A Dose of Perspective

This is an important distinction to be made. There are certainly cases of misinformation surrounding the topic of AI-assisted hacks. In their reports on threat investigations, Sophos maintains a more nuanced view on the matter: AI is providing advantages for both sides in conducting standard actions quicker, but there is no significant proof yet of attacks that use completely new approaches, which might increase as capabilities increase. scworld

The significance of the point cannot be overstated. Contemporary AI-assisted hacks are relying on old vulnerabilities – software with patches not applied, weak authentication passwords, overprivileged cloud user profiles and misconfigurations. AI is simply amplifying the issue, but not providing any new vulnerabilities. It is reassuring, because at least we know how to deal with the problem. It is terrifying, because most companies do not.

AI Is Also Becoming a Target

But there is another front. In the process of developing and deploying artificial intelligence assistants and autonomous agents having access to your emails, files, source codes, and other tools, a vulnerability is created that may be exploited. The prompt injection, where through hidden prompts in a file, website or message the behavior of the agent is redirected, poses the biggest threat here. It is quite powerful to have an autonomous agent with access to your emails and able to send messages.

The security specialists have raised another concern regarding the practice of inserting AI agents into the security operations center. When done carefully, they reduce alert fatigue. Done carelessly, they become yet another privileged system that can be manipulated by the attackers. The experience with other technological waves teaches us this – each added layer of automation needs its own access controls, monitoring and audit trail.

How Defenders Are Fighting Back

That technology works both ways, and the defenses are employing it.

More efficient detection and triage. AI can go through millions of log files, correlate weak signals and find the handful of alerts that matter, work which humans cannot do efficiently.

Automated vulnerability discovery. Machine learning algorithms can search for vulnerabilities within code repositories before anyone attacks them, moving the process of security earlier in the development cycle.

Fast response. Automated playbooks can isolate the compromised machine or cancel its credentials in seconds, not hours.

Improved training. There are already companies that use AI to conduct simulated attacks and train their personnel.

The defender’s edge, if there is one, comes from context. A corporation understands its system, and machine learning algorithms can detect anomalies within the system that is being monitored. The problem is that the defender must win each and every time, but the attacker needs only one success.

What Organizations and Individuals Should Do

For organizations, the priorities are fairly clear:

Patch quickly. Recognize that exploit windows have shrunk and start reducing patch intervals, particularly those involving internet-exposed assets.
Reduce identity and access risks. Enable multi-factor authentication, eliminate excess permissions, and recognize AI agents as identities operating under minimal privileges.
Secure your AI ecosystem. Make sure that all actions performed by the agents are logged, give them as few permissions as possible, and conduct injection tests before putting them into action.
Validate out-of-band. For all payment transactions, credential changes and critical actions, verify through an independent channel regardless of voice/video confirmation.
Get ready to handle a disaster situation. Perform your drills assuming that the attacker will move quicker than you are used to moving.
As far as people are concerned, the advice is simple: be suspicious about urgency, use password managers and passkeys, and come up with a "safe word" to communicate with your family if your voice gets cloned.

The Road Ahead

AI is not the end of cybersecurity, nor is it giving attackers all the upper hand. All it does is decrease the time and costs of attacks, and defenders’ response to them. The organizations that will survive will take AI into account both as a threat and a tool for defense at the same time: utilizing it in their operations while securing themselves from any potential misuses of it.

These recent attacks should be considered as a warning bell rather than a disaster, since time for preparations still remains but it will soon run out according to the experts.

Author:

Rajat Sharma

Related Links:

Anthropic AI Tool

What is Writesonic

Career Objectives For Fresher

Resume Tips For Software Developers

Do visit our channel to know more: SevenMentor