New International AI Safety Report Spotlights Emerging Risks
The International AI Safety Report 2026 found increasing and emerging concerns around the use of artificial intelligence (AI) in deepfakes, biological weapons, and cyberattacks. The authors reviewed the latest scientific research into the risks and abilities of general-purpose AI, which are models and systems that can perform a wide range of tasks in different contexts, such as creating text, images, audio, or other data that can be adapted to specific scenarios or applications.
This is the second edition of the report published by the AI Security Institute (part of the UK Department for Science, Innovation, and Technology) and supported by more than 30 nations and international organizations.
“General-purpose AI systems are already causing real-world harm,” the report said. “…Advances in AI capabilities may pose further risks that have not yet materialized. Understanding these risks, including their mechanisms, severity, and likelihood, is essential for effective risk management and governance.”
The report identified and organized the risks into three categories: risks from misuse, risks from malfunctions, and systemic risks.
Risks from misuse are ones where someone intentionally uses an AI system to cause harm or for criminal purposes. This category included deepfakes, biological or chemical risks, and cyberattacks.
Deepfakes
This can include using AI-generated images, videos, text, or audio for defamation, extortion, fraud, non-consensual intimate images, and child sexual abuse material, according to the report.
“Accessible AI tools have substantially lowered the barrier to creating harmful synthetic content at scale. Many tools are free or low-cost, require no technical expertise, and can be used anonymously,” the report said.
The use of deepfakes, especially amongst fraudsters and scammers, is increasing. Malicious actors can use voice clones and deepfakes to impersonate executives and family members to dupe victims into transferring funds. Criminals have also used this kind of content for identity theft (authorizing bank transfers by impersonating the true owner’s voice or likeness), blackmail, or sabotage.
“Researchers have also noted that deepfakes may risk undermining the reliability of evidence presented in court proceedings,” the report added.
AI-generated sexual content is increasingly prevalent and the reality and complexity of these images has significantly improved. Another concerning abuse of AI is with deepfake pornography, which mostly targets women and girls.
“One study estimated that 96 percent of deepfake videos are pornographic, that 15 percent of UK adults report having seen deepfake pornographic images, and that the vast majority of ‘nudify’ apps explicitly target women,” the report said. “…Sexual deepfakes are also used in intimate partner abuse, again disproportionately affecting women.” Unfortunately, even though many systems have safeguards to prevent this type of abuse, users have found way to either bypass them or use other programs that lack protections.
Biological and Chemical Risks
AI systems are now advanced enough that they can provide sufficiently detailed scientific information and help with specialized laboratory methods. This can mean positive advancements for the scientific community with AI systems generating experimental protocols, troubleshooting technical obstacles, and even designing molecules and proteins for legitimate research, such as accelerating the discovery of new beneficial drugs and medicines and improving disease diagnosis.
But that same advancement can help threat actors create biological and chemical weapons. “By combining and interpreting existing complex information on the Internet that is relevant to weapons development, and tailoring advice to specific malicious activities, AI systems can lower existing expertise barriers, allowing more actors to cause harm,” the report noted. To try to prevent this type of abuse, several major developers released new systems with additional safeguards in 2025.
However, adequately analyzing and determining the impact that AI systems have on biological and chemical risks is its own challenge because the research into the issue is either limited or could itself result in creating harm. “For example, if researches carry out, or publish the results of, a study on AI assistance in weapons development, they may risk inadvertently violating national security laws or treaties such as the Biological Weapons Convention and Chemical Weapons Convention. …Relevant data is also often classified, particularly when it relates to the use of AI systems by state actors,” the report said.
Cyberattacks
While cyberattacks are nothing new, the report found that criminal groups and state-sponsored attackers actively use general-purpose AI systems to carry out or assist with their attacks. Executing a cyberattack involves multiple steps, including reconnaissance, identifying vulnerabilities, developing the attack, attack execution, maintaining persistent access, and carrying out the desired objective.
“Extensive research shows that AI systems can now support attackers at several steps of the ‘cyberattack chain,’” according to the report. “…One area where there is particularly strong evidence that AI systems provide meaningful assistance is in discovering ‘software vulnerabilities’: weaknesses in programs that can be exploited to compromise the security of computer systems.”
These services are for sale, with pre-packaged AI tools and AI-generated ransomware popping up in underground marketplaces.
Other Risks
The report also identified other risks linked to general-purpose AI-systems.
Risks from malfunctions. These are risks that arise from failures of AI systems, such as false of fabricated information (commonly known as hallucinations), producing flawed code, and misleading medical advice. These failures, when left unchecked by users or taken at face value, can result in physical or psychological harm and a user’s exposure to reputational damage, legal liability, or financial loss.
Another risk within this category is the hypothesized loss of control, where at least one AI system operates beyond anyone’s control. Regaining that control can be expensive or even impossible. The report noted that experts’ opinions on the likelihood of this varies greatly, with some calling it implausible while others find it sufficiently likely.
“Loss of control becomes more likely if AI systems are ‘misaligned,’ meaning they have goals that conflict with the intentions of developers, users, or society more broadly,” the report said.
Systemic risks. While AI-systems can be seen and used as a tool to automate and expedite certain functions, those benefits also present risks to the work force and human autonomy. However, predicting these impacts is tricky.
“Around 60 percent of jobs in advanced economies and 40 percent in emerging economies are exposed to general-purpose AI, but the impacts of this will depend on how AI capabilities develop, how quickly workers and firms adopt AI, and how institutions respond,” the report noted.
As for impacts on humans’ autonomy, AI systems can negatively impact this in several ways, including on cognitive skills, how humans develop beliefs and preferences, and how they make and act on decisions. “For example, one study found that three months after the introduction of AI support, clinicians’ ability to detect tumors without AI assistance had dropped by 6 percent,” according to the report.
