Current AI models are capable of causing cyberattack, bioterrorism and psychosocial risks.

Judge as of 2026-08-26. The claim names three risk types; each holds if credible evidence shows that current AI models can materially contribute to it. That evidence can come from lab system cards, government or safety-institute assessments, or documented incidents: AI-assisted cyberattacks; meaningful uplift toward biological weapons, as rated by lab or government evaluations; and psychological or social harms such as dependency, manipulation or mental-health harms linked to chatbots. True if all three are supported. Mostly true if all three are supported but one rests on weaker or more contested evidence. Mixed if two of the three are supported. Mostly false if only one is supported, or if the claims rest on speculative capability rather than evidence. False if evidence shows current models cannot meaningfully contribute to any of them.

Set before the research began.

1. E1 Hackers used autonomous AI agents to attack Taiwan. Is this the future of cyberwarfare? | CNN Business CNN · 13 August 2026

   “In this case, an autonomous system coordinating up to eight AI agents – carrying out intrusions and deciding what to do next without human intervention – effectively ran a hacking campaign rather than just merely assisting a human hacker.”

2. E2 Hackers used autonomous AI agents to attack Taiwan. Is this the future of cyberwarfare? | CNN Business CNN · 13 August 2026

   “The investigation found clear indications that the attacks originated overseas and involved a hybrid approach in which hackers combined conventional operations with AI agents such as OpenClaw,”

3. E3 International AI Safety Report 2026 International AI Safety Report (chaired by Prof. Yoshua Bengio; UK Government publication, DSIT research series) · February 2026

   “Major AI developers have released (some) recent models with heightened safeguards after being unable to exclude the possibility that they could meaningfully assist novices in creating biological weapons.”

4. E4 International AI Safety Report 2026 International AI Safety Report (chaired by Prof. Yoshua Bengio; UK Government publication, DSIT research series) · February 2026

   “AI systems now match or exceed expert performance on many benchmarks measuring knowledge relevant for biological weapons development.”

5. E5 International AI Safety Report 2026 International AI Safety Report (chaired by Prof. Yoshua Bengio; UK Government publication, DSIT research series) · February 2026

   “However, substantial uncertainty remains about how these capabilities affect risk in practice, given material barriers to weapons production and the difficulty of conducting uplift studies”

6. E6 GPT-5 System Card OpenAI · 7 August 2025 Own record

   “While we do not have definitive evidence that this model could meaningfully help a novice to create severe biological harm”

E1 and E2 report a documented cyber campaign in which autonomous AI agents carried out intrusions, so the cyberattack risk is supported. On biological risk, the International AI Safety Report says AI systems match or exceed experts on bioweapons-relevant knowledge benchmarks (E4), and that developers added safeguards because they could not rule out meaningful help to novices (E3). The same report notes substantial uncertainty about the real-world effect (E5), so this support is weaker and more contested. The evidence provided says nothing on psychosocial harms, so two of the three risk types are supported. That fits the criteria for mixed.