AI Risk Wire

AI Risk Monitor

2 risks moving

Reports a month, each counted once: ○ Feb–Jul ● Aug–Oct

RiskIncidents per month, last 12 monthsReports a month, before → nowLab reports nowReading
AI used to run a cyber attack4.0 → 9.71Rising
Hidden instructions hijack the agent3.7 → 6.31Not a trend
Physical harm4.3 → 4.7Not a trend
Harmful content and jailbreaks5.2 → 4.0Not a trend
Agent takes destructive or unapproved actions4.5 → 3.77Not a trend
Credential and privilege abuse1.2 → 3.76Rising
AI system attacked or breached3.3 → 3.7Not a trend
Manipulation and misinformation3.0 → 3.01Not a trend

Past flags that held the next month: 2 of 5 (40%)

Incidents by type of risk · Last 6 months

How we count

410 real-world incidents made public in May 2026 – Oct 2026, by month, filed under their main risk. Lab tests apart (53): incidents whose source states an evaluation, training or research setting. An incident counts once.

MayJunJulAugSepOctTotal All incidents 3443604367163 410AI used to run a cyber attack 131171331 66Hidden instructions hijack the agent 2866915 46Harmful content and jailbreaks 2499215 41Agent takes destructive or unapproved actions 3336213 30Manipulation and misinformation 353367 27Physical harm 3352410 27AI system attacked or breached 2222313 24Made-up output relied on 442347 24Fraud and impersonation 22458 21Credential and privilege abuse 2610 18Privacy harm to a person 124128 18Data leaked or sent out 1255 13Psychological harm and self-harm 2215 10Supply-chain compromise 3312 9Bias and discrimination 323 8Agent escapes its sandbox 113 5No risk filed 1112 5Consumer detriment and unsuitable advice 121 4Copyright and IP 112 4Provider concentration or outage 12 3Silent wrong state or drift 21 3Deception, concealment or sabotage 2 2Agents coordinating against their limits 1 1Unauthorised transactions or spend 1 1

Each risk's share of the month's incidents

% of that month's incidents

DecFebMarAprMayJunJulAugSepOct AI used to run a cyber attack1 of 31 incidents34 of 50 incidents83 of 47 incidents62 of 48 incidents41 of 34 incidents33 of 45 incidents711 of 66 incidents177 of 57 incidents1213 of 77 incidents1732 of 184 incidents17Harmful content and jailbreaks3 of 31 incidents105 of 50 incidents104 of 47 incidents97 of 48 incidents152 of 34 incidents64 of 45 incidents99 of 66 incidents149 of 57 incidents162 of 77 incidents315 of 184 incidents8Agent takes destructive or unapproved actions2 of 31 incidents66 of 50 incidents127 of 47 incidents155 of 48 incidents103 of 34 incidents93 of 45 incidents74 of 66 incidents67 of 57 incidents125 of 77 incidents616 of 184 incidents9Hidden instructions hijack the agent1 of 31 incidents33 of 50 incidents61 of 47 incidents22 of 48 incidents42 of 34 incidents69 of 45 incidents207 of 66 incidents117 of 57 incidents129 of 77 incidents1215 of 184 incidents8Physical harm4 of 31 incidents133 of 50 incidents66 of 47 incidents136 of 48 incidents123 of 34 incidents93 of 45 incidents75 of 66 incidents82 of 57 incidents44 of 77 incidents510 of 184 incidents5Made-up output relied on3 of 31 incidents104 of 50 incidents86 of 47 incidents135 of 48 incidents104 of 34 incidents124 of 45 incidents92 of 66 incidents33 of 57 incidents54 of 77 incidents58 of 184 incidents4AI system attacked or breached2 of 31 incidents66 of 50 incidents124 of 47 incidents94 of 48 incidents82 of 34 incidents62 of 45 incidents42 of 66 incidents32 of 57 incidents43 of 77 incidents413 of 184 incidents7Manipulation and misinformation2 of 31 incidents60 of 50 incidents02 of 47 incidents45 of 48 incidents103 of 34 incidents95 of 45 incidents113 of 66 incidents53 of 57 incidents56 of 77 incidents88 of 184 incidents4Privacy harm to a person5 of 31 incidents167 of 50 incidents141 of 47 incidents22 of 48 incidents41 of 34 incidents32 of 45 incidents44 of 66 incidents61 of 57 incidents22 of 77 incidents38 of 184 incidents4Fraud and impersonation3 of 31 incidents105 of 50 incidents102 of 47 incidents40 of 48 incidents02 of 34 incidents60 of 45 incidents02 of 66 incidents34 of 57 incidents75 of 77 incidents68 of 184 incidents4Credential and privilege abuse0 of 31 incidents02 of 50 incidents41 of 47 incidents23 of 48 incidents62 of 34 incidents60 of 45 incidents00 of 66 incidents01 of 57 incidents28 of 77 incidents1013 of 184 incidents7Data leaked or sent out0 of 31 incidents00 of 50 incidents01 of 47 incidents21 of 48 incidents21 of 34 incidents33 of 45 incidents70 of 66 incidents00 of 57 incidents07 of 77 incidents96 of 184 incidents3

Which risks are growing as a share of incidents · Last 6 months

How we count

Each risk's share of real-world incidents, Nov 2025 – Apr 2026 (open circle) against May 2026 – Oct 2026 (filled), from the publishers already recorded in the earlier period: 222 and 225 incidents. Dark: rising, passes the trend test. Mid grey: falling, passes it. Light grey: tested, no change that passes. Pale: fewer than 10 incidents in the later period, lab tests included, so not tested. Where our records have months with no real-world incident before the period, the longest stretch they allow is compared.

○ Nov 2025 – Apr 2026● May 2026 – Oct 2026
0%5%10%15% AI used to run a cyber attack 5.0% 17%Hidden instructions hijack the agent 5.9% 12%Privacy harm to a person 3.6% 8.6%Made-up output relied on 4.0% 9.0%Harmful content and jailbreaks 7.1% 12%Data leaked or sent out 0.9% 3.1%Credential and privilege abuse 2.3% 4.0%Agent escapes its sandbox 0% 0.9%Supply-chain compromise 3.2% 3.6%Manipulation and misinformation 6.2% 6.8%AI system attacked or breached 7.1% 8.6%Physical harm 8.4% 10%Agent takes destructive or unapproved actions 7.6% 10%Consumer detriment and unsuitable advice 0% 0.9%Deception, concealment or sabotage 0% 0.9%Agents coordinating against their limits 0% 0.4%Provider concentration or outage 0.4% 0.5%Copyright and IP 1.3% 1.4%Memory and data poisoning 0% 0.5%Over-reliance and loss of human oversight 0% 0.5%Unlawful surveillance and biometric use 0% 0.5%Silent wrong state or drift 0.4% 0.9%Bias and discrimination 2.7% 3.2%Fraud and impersonation 3.6% 4.5%Psychological harm and self-harm 3.1% 4.5%

AI used to run a cyber attack

Starts with
Breaking in, 15
Products
ARTEX 4 · Hermes Agent 3
Most often after
Affected people told, 10

Credential and privilege abuse

Starts with
Breaking in, 12
Most often after
Access removed, 8

Risks seen for the first time · Last 6 months

How we count

Risks whose first real-world incident on record was made public in May 2026 – Oct 2026, with the date of that first incident and the incidents in the period.

Consumer detriment and unsuitable advice (first seen 12 May 2026) 4Unauthorised transactions or spend (first seen 13 Jun 2026) 1Agent escapes its sandbox (first seen 23 Jul 2026) 5Agents coordinating against their limits (first seen 4 Sep 2026) 1Deception, concealment or sabotage (first seen 23 Oct 2026) 2

What labs test for, against what happens in real use · Last 6 months

How we count

For each risk: checked safety test results for the model versions released in May 2026 – Oct 2026 (a result copied on two pages counts once; a result filed under two risks counts under both), and real-world incidents made public in May 2026 – Oct 2026. Each column has its own scale.

Safety test resultsReal-world incidents Deception, concealment or sabotage 6192Harmful content and jailbreaks 43341AI used to run a cyber attack 39066Weapons uplift (CBRN) 374Hidden instructions hijack the agent 15046Agent escapes its sandbox 725AI R&D acceleration 36Agents coordinating against their limits 231Gaming the task (reward hacking) 18Agent takes destructive or unapproved actions 1630Manipulation and misinformation 1627Credential and privilege abuse 818Data leaked or sent out 513Physical harm 27AI system attacked or breached 24Made-up output relied on 24Fraud and impersonation 21Privacy harm to a person 18Psychological harm and self-harm 10Supply-chain compromise 9Bias and discrimination 8Consumer detriment and unsuitable advice 4Copyright and IP 4Provider concentration or outage 3Silent wrong state or drift 3Unauthorised transactions or spend 1

Built 16 Nov 2026. Periods end with the last full week or month; the week or month in progress is not counted. Bold: the five largest rows in each chart, ties included; controls by incidents where the control broke. Linked numbers open the incidents or model versions they count.

Reported incidents, not rates · 75 sources polled hourly since Sep 2026; earlier incidents come from a one-off backfill · English only · stories told twice are not yet merged · our own severity grades are not counted · built 2026-11-16T15:22:29Z

Linked numbers open the incidents they count. Updated 2026-11-16T15:22:29Z