AI Cybersecurity Attacks: How Hackers Are Using AI
.webp "AI Cybersecurity Attacks: How Hackers Are Using AI") |
| AI Cybersecurity Attacks: How Hackers Are Using AI |
AI is no longer being used only to help attackers write phishing emails or generate code. Security researchers have documented cases where AI systems have been used to automate reconnaissance, vulnerability discovery, exploitation and other parts of cyberattacks with limited human involvement. This matters because AI can make some attacks faster, easier to scale and less dependent on constant human direction. This article looks at documented cases of AI being used in cyberattacks and what they mean for your security.
Why 2026 Is Different
For years, attackers mainly used AI as a helper for tasks such as writing phishing messages, translating text or generating code. That picture is changing. In a 2025 campaign that Anthropic attributed to a Chinese state-sponsored group called GTG-1002, attackers used Claude Code as an autonomous operator for reconnaissance, vulnerability discovery, exploitation, credential harvesting and data analysis. Anthropic estimated that AI handled 80–90% of the tactical work, while human operators still selected targets and directed the overall operation. IBM's 2026 research also found that AI-enabled attacks are becoming more significant, while the FBI's 2025 Internet Crime Report recorded more than 22,000 complaints involving AI-related fraud and adjusted losses of more than $893 million. These findings do not mean most cyberattacks are autonomous, but they show that AI is becoming an increasingly useful part of the attack process.
Six Ways Hackers Are Using AI Right Now
1. AI-Written Phishing and Scam Messages
The oldest AI-assisted attack is still one of the effective ways to trick people. In the past you could spot an email because the writing was bad and full of mistakes. Now AI writes messages that sound very natural and personal. These AI messages can talk about your job, your coworkers or things you did recently. AI often gets this information from media or company websites. This makes phishing harder to see. You cannot look for grammar mistakes or generic greetings anymore because the AI makes the phishing messages look real.
2. Voice Cloning and Deepfake Calls
Voice cloning is another growing tool in fraud and social engineering. Attackers can use publicly available audio to create convincing impersonations and then use those voices in calls or messages designed to pressure victims into sending money or revealing sensitive information. The FBI has warned that criminals are using generative AI to create convincing voice and video impersonations as part of fraud schemes.
These scams often follow familiar social-engineering patterns with AI making the impersonation more convincing. A fake voice may claim that a relative needs money urgently, while a deepfake video or voice may impersonate a company executive and pressure an employee to transfer funds. In one widely reported case in Hong Kong, an employee was tricked into transferring about $25 million after joining a video conference in which the other participants appeared to be company executives but were deepfakes.
3. Fully Autonomous Hacking Agents
This is one of the more concerning developments because AI agents can perform multiple steps of an attack without a person directing every action. In May 2026, Sysdig documented an intrusion in which an LLM agent performed post-compromise actions in real time, moving from a compromised application to an internal PostgreSQL database. The full attack chain ran end-to-end in under an hour, showing how an AI agent can adapt its actions as it receives new information
4. AI-Assisted Malware and Vulnerability Discovery
In May 2026, Google Threat Intelligence Group reported that it had identified an attacker using a zero-day exploit that researchers believed had been developed with AI assistance. Google said the threat actor planned to use the exploit in a wider attack, but the discovery happened before that campaign could be carried out at scale. The case shows how AI can potentially speed up vulnerability research and exploit development, although it does not mean AI can automatically discover and exploit every software flaw.
5. AI-Powered Password and Credential Attacks
AI can help attackers automate and personalize credential attacks, including generating likely passwords or identifying useful information about a target. It can also help attackers process stolen credentials and prioritize targets more efficiently. Reused passwords remain a major risk because a password exposed in one breach can potentially be used to access another account.
6. Chatbot-Driven Social Engineering
Not every AI‑enabled attack needs technology. I notice that some of the damaging recent breaches involve attackers calling company help desks and convincing staff to reset passwords or bypass multi‑factor authentication. AI can help write the conversation or produce supporting details. Security researchers say this does not require technical skill, only persistence and a good script. That is exactly the type of task AI can produce on demand.
Real Documented Cases From 2026
A few cases stand out because they were independently verified than just predicted:
- The GTG-1002 espionage campaign (disrupted November 2025 reported publicly by Anthropic): A state-linked group used an AI coding tool to perform most of a cyberespionage operation against government and corporate targets with very little human input.
- The Thailand Ministry of Finance intrusion (mid-to- June 2026): An autonomous AI agent broke in mapped the network and destroyed data using a generated encryption key that was never saved anywhere. The damage was permanent even if a ransom had been paid.
- The Taiwan government attack (reported August 2026): An autonomous AI system was used in what researchers call one of the fully AI-run attacks, on a national government.
- Googles May 2026 zero-day discovery report: Threat researchers identified an AI-developed exploit before it could be used in a planned mass-exploitation campaign.
These cases share a pattern. A human still chooses the target. Sets up the operation but the AI agent does the technical work: scanning, exploiting, adapting and sometimes destroying evidence. The AI works at a speed and scale that humans cannot match on their own.
Why This Affects More Than Big Companies
It is easy to think that this danger only affects governments and big companies. That is not the whole story. AI has lowered the skill needed to carry out an attack. Security researchers say that some recent high-profile breaches were carried out by attackers with basic technical skills, who used AI and social engineering instead of deep hacking knowledge and caused damage similar to that of more sophisticated groups.
On the consumer side, voice-cloning scams can target families directly. Attackers may impersonate relatives, friends, executives or other trusted people and use urgency to pressure victims into sending money or revealing information. The tools are becoming easier to access, which lowers the barrier for criminals who want to run convincing impersonation scams.
How to Protect Yourself and Your Business
You can't stop AI-powered attacks from existing, but you can make yourself a much harder target.
For individuals:
- Set up a family codeword that only real family members know, and use it to verify unexpected calls asking for money.
- Never act on an urgent money request from a call or voicemail alone. Hang up and call the person back on a number you already have saved.
- Use a password manager and unique passwords for every account, since AI-assisted credential attacks rely heavily on reused passwords.
- Turn on multi-factor authentication, ideally using an authenticator app or hardware security key rather than text messages, which can be intercepted.
- Be cautious about how much personal audio and video you post publicly. Publicly available voice recordings can potentially be used to create convincing voice impersonations, so limiting unnecessary exposure can reduce the material available to scammers.
For businesses:
- Require a second verification step, through a separate channel, for any request involving a wire transfer or sensitive data, no matter how urgent the request sounds.
- Train help desk and support staff to verify identity carefully before resetting passwords or MFA, since this has become a favorite target for social engineering.
- If your company uses AI agents internally, limit what systems and data they can access, monitor their activity, and require human approval before any agent can take a high-risk action like moving money or deleting data.
- Keep software patched. Several documented AI-driven attacks succeeded through known vulnerabilities that already had available fixes.
Common Questions About AI Cyberattacks
Can I tell if I'm talking to a deepfake or a cloned voice? Not reliably. High-quality synthetic audio and video can be difficult to identify by appearance or sound alone. Instead of trying to detect a fake yourself, verify unexpected requests through a separate trusted channel, such as calling a known phone number or confirming the request directly with the person.
Is AI making all hackers more dangerous or only a few advanced groups? Both effects are possible. The most sophisticated autonomous operations documented by Anthropic and security researchers involve capable threat actors, while simpler AI-assisted phishing, impersonation and social-engineering scams require much less technical expertise. AI therefore has the potential to help both highly skilled attackers and less experienced criminals..
Does antivirus software still work against attacks that use AI? Yes. Antivirus, endpoint protection, firewalls and other security controls remain useful even when attackers use AI. They are not designed to detect “AI” itself; they look for malicious behavior, suspicious files, network activity and other indicators of compromise. Because AI-assisted attacks can also involve phishing or social engineering without malware, organizations should use multiple layers of protection rather than relying on one security product.
Should I be worried about AI agents, not about AI chatbots? Yes this is something that should be watched closely. The difference between a chatbot that gives answers and an agent that can take actions, like running commands searching the web or looking at files is big. Attacks that use AI agents happen faster. Need less continuous help from people, which is one of the reasons why security experts are considering agent behavior as a separate kind of risk instead of grouping it with older types of AI misuse.
The Bottom Line
AI is changing parts of cybercrime by helping attackers automate tasks, process information faster and create more convincing social-engineering content. Documented cases from 2025 and 2026 show that some threat actors are using AI agents for substantial parts of cyber operations, while other criminals are using AI for simpler phishing and impersonation scams. The most advanced autonomous attacks remain less common than ordinary AI-assisted attacks, but the underlying tools are becoming more accessible. For individuals, the practical response is still straightforward: use strong unique passwords, enable MFA, verify unusual requests through a separate channel and keep devices and software updated.
The best way to protect yourself isn't a piece of software. It's a habit: slow down check through a method and look at urgent requests for money or access, with doubt no matter how real they seem or look.
Sources
- Anthropic — Disrupting the First Reported AI-Orchestrated Cyber Espionage Campaign
- Google Threat Intelligence Group — AI-Related Threat Activity and Zero-Day Exploit Findings, May 2026
- Google Threat Intelligence Group — The Evolution of Adversarial AI, September 2026
- IBM — Link to the specific 2026 research report supporting the article's AI-enabled breach claims.
- FBI Internet Crime Complaint Center (IC3) — 2025 Internet Crime Report
- Sysdig Threat Research Team — How an Attacker Used an AI Agent to Reach an Internal Database
Disclaimer: This article is intended for general educational and informational purposes. This article does not provide instructions for carrying out attacks. Is not a substitute for professional cybersecurity or legal advice. Threat data changes quickly. Readers should verify statistics and guidance through official sources such, as CISA, the FBI and the security vendors cited above before making security decisions for their organization.