AI is helping threat actors do familiar cyber work faster and at greater volume; the evidence does not show that it has broadly invented a new kind of attack. Google’s analysis of government-backed actors using Gemini found assistance with tasks across the attack lifecycle, but no indication that the observed activity was developing novel capabilities. That is meaningful help—not proof of autonomous, end-to-end compromise.
What threat actors have used generative AI to do
Google Threat Intelligence Group’s January 29, 2025 report examined government-backed actors’ interactions with Gemini. Its findings are specific to that activity, not a survey of every model or threat actor. GTIG used analyst review and LLM-assisted analysis.
As an Amazon Associate I earn from qualifying purchases.
The report describes AI as a way to support work actors already do, including:
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errors- Researching targets, infrastructure, and reconnaissance findings.
- Investigating vulnerabilities and developing payloads.
- Writing scripts and seeking help with evasion.
- Drafting, translating, or otherwise working with content.
These examples span stages of an operation, but assistance at multiple stages is not the same as an AI system independently carrying out an intrusion. GTIG’s conclusion was: “Rather than enabling disruptive change, generative AI allows threat actors to move faster and at higher volume.” (Google Cloud / GTIG, “Adversarial Misuse of Generative AI,” January 29, 2025.)
#1 Best Overall
What AI changes for attackers—and what the evidence does not show
It reduces friction in routine work
Generating or adapting text, getting coding help, and organizing research can take less effort when an AI tool assists. Repeating those tasks at higher volume may also become easier. GTIG describes AI as potentially useful to skilled actors as a productivity framework and to less skilled actors as a learning aid for developing knowledge and tools. That is the report’s assessment, not a measured effect that applies uniformly to every actor.
It does not establish a new attack playbook
GTIG reported no indications of novel capabilities in the government-backed actors’ observed Gemini activity. This does not prove that no one has ever used AI in a novel way; it sets a boundary on what this report’s evidence supports. The observed examples show help with familiar tasks, not proof that generative AI has made phishing, exploitation, or intrusion methods fundamentally new.
It does not establish a universal speed or success rate
The reviewed reports provide no general statistic for how much AI shortens an attacker’s timeline or increases the likelihood of a successful attack. “Faster and at higher volume” is GTIG’s qualitative conclusion, not a percentage or guaranteed result for a particular organization.
Observed assistance is different from projected agent automation
Microsoft’s Digital Defense Report 2025 describes AI agents as having the potential to automate a broader attack lifecycle, including reconnaissance, vulnerability scanning, and exploitation at scale. The important distinction is that this is a stated possibility. It should not be recast as evidence that autonomous agents are already carrying out end-to-end attacks across the board.
Rank #3
Microsoft also describes AI as dual-use: defenders can apply it to threat-intelligence analysis, finding protection gaps, and automating responses, while attackers can target insecure AI workloads and use synthetic media for fraud. As Microsoft puts it, “Both adversaries and defenders are using AI to make their operations more effective and efficient, rendering the technology a cybersecurity risk and tool at once.” The report describes possible uses on both sides; it does not establish that defensive gains consistently outweigh attacker gains.
Why conventional attack-path defense still matters
AI assistance does not remove the need to secure the routes into valuable systems. Microsoft’s 2024 Digital Defense Report explains that attack-path analysis can combine asset inventories, vulnerability data, and external attack surfaces to identify possible chains leading to critical assets. Its June 2024 infographic reported the following results from Microsoft Security Exposure Management’s analysis:
| Finding | Microsoft-reported figure |
|---|---|
| Organizations exposed to at least one attack path | 90% |
| Attack paths leading to a sensitive user account | 61% |
| Organizations with attack paths exposing critical assets | 80% |
| Attack paths including lateral movement based on non-interactive remote code execution | 40% |
| Attack paths containing three steps or fewer | 10% |
| Organizations exposed to more than 1,000 attack paths | 3% |
These are figures from Microsoft’s analysis, not universal rates for all organizations. They illustrate why an organization should look for reachable paths and weak controls rather than assume an attack must be technically novel to be dangerous. (Source: Microsoft, Digital Defense Report 2024, infographic titled June 2024.)
Recommended Free Tools
Practical steps for security teams
Focus on the weaknesses that could let a familiar technique reach a critical asset, while governing AI used in both security work and business systems.
Best Value
- Map likely paths to critical assets. Keep asset inventories current and review vulnerabilities and exposed services to identify plausible routes into important systems.
- Address outdated controls and technical debt. Prioritize fixes based on whether a weakness creates a practical path to a sensitive account or critical asset.
- Find and govern shadow IT. Identify unapproved tools and workloads, including AI services that may handle organizational data or connect to internal systems.
- Update data-security policies. Define what information staff may submit to AI tools and how outputs that influence security decisions must be checked.
- Validate AI-assisted defense. Use AI for analysis or response automation only with appropriate review, permissions, and safeguards; a claimed capability is not a substitute for verifying the result.
Microsoft’s recommendations in its 2024 report include addressing technical debt, outdated controls, and shadow IT, and identifying likely attack paths to critical assets. Its 2025 report’s defensive AI examples add tools for threat analysis, gap identification, and response automation, which organizations still need to govern and validate.
The useful distinction
For now, the strongest evidence here supports a practical conclusion: AI can lower the effort needed for familiar cyber tasks and help actors scale them, while the broader prospect of agentic, end-to-end automation remains a possibility rather than a result established by the cited observations. Defenders should prepare for faster, more assisted work without treating every AI-assisted incident as a new class of attack.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
