Table of Contents
Adversaries are using AI to sharpen attacks, automate operations, and challenge long-standing defenses, according to a new Microsoft report. Researchers describe a year in which criminal and state-backed actors blurred the lines between cybercrime, espionage, and disruption, targeting public and private sectors.
Identity at the center of attacks
The report shows that most breaches begin with stolen or guessed passwords. More than 97% of identity-based attacks rely on password spray or brute-force techniques. While MFA blocks most attempts, many organizations have coverage gaps, particularly for service accounts and non-human identities.
In the first half of 2025, identity-based attacks rose by 32%. Research and academia were hit hardest, driven by open networks and decentralized IT systems. Attackers increasingly target workload identities, applications, scripts, and cloud services that hold elevated privileges but are often less protected than user accounts.
Criminal groups focus less on breaking in and instead use stolen credentials. Infostealer malware such as Lumma and RedLine collect credentials later sold on dark web markets. Access brokers resell this information to ransomware and data extortion groups. Intel 471 identified 368 active brokers in the past year, affecting victims in over 130 countries.
Ransomware finds new leverage in hybrid systems
Attackers are exploiting exposure across the cloud and supply chains. A third of incidents stemmed from neglected weaknesses such as unpatched web assets, exposed remote services, and misconfigured perimeter systems. Nearly 18% of attacks started with vulnerable web applications, and 12% involved remote access services.
Researchers observed an 87% rise in destructive actions in cloud environments, including mass deletions and ransomware. Hybrid operations now account for more than 40% of ransomware incidents, up from less than 5% two years ago. Attackers are exploiting misconfigurations and weak access controls to move between cloud and on-premises systems.
Ransomware continues to be the most common and costly threat. More than half of attacks with known motives were financially driven, while only a small share involved espionage.
Operators rely less on phishing and turn to social engineering. Voice-based “help desk” scams and impersonation through collaboration tools like Teams have become common entry points. Attackers also use legitimate remote monitoring tools to stay hidden and maintain access.
State actors turn to AI for cyber operations
Attackers use AI to write phishing messages, find system weaknesses, and adjust malware behavior. These tools reduce human effort and make large-scale operations easier to run.
At the same time, AI systems themselves are under attack. Criminals use prompt injection and data manipulation to alter model behavior or gain access to information. As organizations adopt AI tools, many overlook the new entry points these systems can create.
“We’re witnessing adversaries deploy generative AI for a variety of activities, including scaling social engineering, automating lateral movement, engaging in vulnerability discovery, and even real-time evasion of security controls. Autonomous malware and AI-powered agents are now capable of adapting their tactics on the fly, challenging defenders to move beyond static detection and embrace behavior-based, anticipatory defense,” said Amy Hogan-Burney, Corporate VP, Customer Security & Trust, and Igor Tsyganskiy, Corporate VP and CISO at Microsoft.
State-backed groups continue to increase their activity. China, Iran, Russia, and North Korea remain active, with campaigns focused on espionage, influence operations, and disruption. This year saw heavier targeting of communications, research, and academia, sectors rich in intellectual property and global connections.
Nation states are also using AI to scale operations and improve precision. Influence campaigns now feature synthetic media and deepfakes to shape narratives in real time. Governments and private actors are collaborating in offensive operations, complicating attribution and response.
The United States accounted for nearly a quarter of observed cyber incidents in early 2025, followed by the United Kingdom, Israel, and Germany. Government agencies and IT companies were the most targeted sectors, each representing about 17% of global attacks. Research, academia, and nongovernmental organizations followed closely.
These patterns show that adversaries continue to target entities holding sensitive data or running essential services. Local governments, often dependent on legacy systems and limited security teams, remain exposed.
