Is artificial intelligence turning every criminal into an expert hacker? During 2026, three of the companies best placed to know — Google, Anthropic and Microsoft — each published research on how attackers use AI. Read together, they give a clear and sober answer: AI is making attacks faster, cheaper and more automated, but most still succeed through weaknesses that organisations already know how to fix.
February: Google sees experimentation, not breakthroughs
In its AI Threat Tracker of February 2026, covering the last quarter of 2025, the Google Threat Intelligence Group (GTIG) reported that it had not yet observed state-backed hacking groups or influence operations achieving breakthrough capabilities that fundamentally alter the threat landscape.
It did see clear signs of where things are heading:
- AI-assisted malware: a malware family GTIG calls HONESTCUE sends a prompt to Google's Gemini API, receives program code in response, and then compiles and runs it in memory.
- Agentic ambitions: threat actors are beginning to show interest in building AI agents to support malware and tool development.
- Copying models: attempts to extract or "distil" AI models are increasing, with one campaign using more than 100,000 prompts.
- Fake "criminal AI": an underground service marketed as a custom AI was in fact powered by several commercial AI products.
The actors covered included groups from North Korea, Iran, China and Russia.
September: Anthropic documents AI agents doing the work
Anthropic's report Detecting and countering misuse of AI: September 2026, covering activity it disrupted between December 2025 and August 2026, went further. It found that a majority of the operations it described were enabled by AI, either carrying out tasks directly or orchestrating them.
Among the cases:
- breaches completed in two to three hours; in one, a single stolen developer token led to full administrative control of a cloud environment in about three hours;
- an espionage actor that, according to Anthropic, used AI at every stage and, when security tools detected its malware, had AI agents modify and redeploy it;
- a group running "agent swarms", in which a lead agent hands tasks to many sub-agents working in parallel, against around fifty organisations.
The report's central message is that sophistication is no longer the barrier it was: what distinguishes dangerous actors now is their intent rather than their skill.
October: Microsoft on speed and scale
Microsoft's 2026 Digital Defense Report, covering July 2025 to June 2026, said threat actors are incorporating AI into reconnaissance, social engineering, malware and exploit development, and that AI can give them greater speed, scale and ability to tailor their activity. It also stressed the other side: AI is giving defenders new ways to find and fix weaknesses.
According to Help Net Security's coverage of the full report, Microsoft found that the median time from a vulnerability being discovered in the wild to being weaponised was well below 24 hours, and that the most complex intrusions are still largely driven by people rather than machines.
What the three reports agree on
- Attacks are getting faster. Hours, not days, from first access to serious damage.
- Automation lowers the bar. Less-skilled attackers can run operations that once needed expertise.
- The entry points are familiar. Stolen tokens and passwords, unpatched systems and social engineering remain the main ways in.
What it means for you
At home: AI makes scam messages and calls more convincing. Be suspicious of urgency, verify requests for money or codes through a channel you already trust, and use 2-step verification.
If you run a small business: speed is the new threat, so make the basics automatic: automatic updates, MFA on every account and backups you can restore. Decide which AI tools staff may use and with which accounts.
If you work in IT or at an MSP:
- Shrink patch times for internet-facing systems — exploitation can now follow disclosure within a day.
- Protect tokens and sessions as well as passwords: short token lifetimes, conditional access and alerts on unusual cloud admin activity.
- Automate detection and containment so that a three-hour breach is stopped in minutes.
- Use AI defensively too: for log triage, phishing analysis and finding misconfigurations.
The bottom line
The 2026 research does not describe unstoppable AI super-hackers. It describes ordinary attacks, carried out faster and at greater scale. Organisations that patch quickly, protect identities and can respond in minutes take away most of the advantage AI gives attackers.
This article summarises and comments on the reports listed below. It is general information, not legal or professional advice.