Harvard Belfer Paper Warns First Major AI-Enabled Terror Attack Could Come by 2027-2028

On October 15, 2026, Harvard Kennedy School's Belfer Center officially released the paper "A Pre-mortem on The First Major AI-Enabled Terrorism Event," authored by Stephen Casper, Max Kamachee, Jordi Weinstock, and others, with team backgrounds spanning MATS and the Harvard Belfer Center. The paper argues that the first major AI-enabled terrorism event could occur as early as 2027 or 2028, and the authors describe it as possibly "the most important paper they have ever written."

Confirmed

• The paper focuses on terrorism scenarios involving AI-assisted cyber, biological, and chemical attack vectors, and offers 9 falsifiable predictions.
• From an AI safety science perspective, it breaks down harmful events into five stages: how harmful capabilities are developed, how they are deployed, how dangerous capabilities can be elicited, how guardrails fail, and how detection fails.
• Harvard Gazette ran an in-depth report on the paper, opening with a fictional narrative involving companies such as OpenAI, Anthropic, Meta, and Google.
• The authors stress that the threat is not purely hypothetical, and note that the paper also includes mitigation recommendations.

Why it matters

• The paper's core conclusion: it is not yet too late to avert catastrophic AI risks, but there are concrete reasons for concern unless AI companies act decisively—providing a clear time-window argument for corporate security investment.
• The pre-mortem format combined with falsifiable predictions means the safety claims can be verified or overturned by future facts, a rare standardized attempt within AI safety research.

2026-10-20 ~ 2026-10-21 · 11 related posts

• [source] New paper warns of first major AI-enabled terrorism event as early as 2027-2028 — StephenLCasper · 2026-10-20
• The paper's five-stage safety analysis: from capability development to detection failure — StephenLCasper · 2026-10-20
• Pathway breakdown of the paper's 9 predictions — StephenLCasper · 2026-10-20
• Paper reviews years of AI misuse: NCII, cybercrime, small-scale terrorism — StephenLCasper · 2026-10-20
• Belfer paper adds 8 more predictions across closed- and open-weight harm pathways — StephenLCasper · 2026-10-20
• Two grounded scenarios: closed-weight bioweapon and open-weight cyberattack — StephenLCasper · 2026-10-20
• [source] Belfer Center publishes pre-mortem paper on first major AI terrorism event with 9 predictions — StephenLCasper · 2026-10-20
• [source] Harvard Gazette: it's not too late to cut catastrophic AI risk — but firms must act now — StephenLCasper · 2026-10-20
• New paper warns of major AI-enabled terrorism event as early as 2027-2028 — StephenLCasper · 2026-10-20

2 near-duplicate retellings: StephenLCasper · StephenLCasper