# Google reports new ways threat actors are using artificial intelligence

> Google Threat Intelligence Group said attackers are targeting AI workflows and cloud resources, including a second-quarter 2026 campaign that moved from cloud access to mass credential harvesting in under six hours.

- Canonical URL: https://newsverum.com/en/news/google-threat-intelligence-adversarial-ai-report-2026
- Publisher: Newsverum
- Section: Security
- Published: 2026-10-06T08:59:00+00:00
- Updated: 2026-10-07T05:57:46+03:00
- Source review period: 2026
- Other editions: [Türkçe](https://newsverum.com/tr/news/google-threat-intelligence-adversarial-ai-report-2026.md), [العربية](https://newsverum.com/ar/news/google-threat-intelligence-adversarial-ai-report-2026.md), [Français](https://newsverum.com/fr/news/google-threat-intelligence-adversarial-ai-report-2026.md), [Español](https://newsverum.com/es/news/google-threat-intelligence-adversarial-ai-report-2026.md), [Deutsch](https://newsverum.com/de/news/google-threat-intelligence-adversarial-ai-report-2026.md), [Português](https://newsverum.com/pt/news/google-threat-intelligence-adversarial-ai-report-2026.md)

*The source does not state the event date; the period shown is the source review period.*

Google Threat Intelligence Group published its assessment of adversarial AI activity on 8 September 2026. It said that in the second quarter of 2026 it observed an actor compromise a cloud resource, then plan and execute an agent-enabled mass credential-harvesting campaign in under six hours.

The report also described theft of AI accounts and API credentials, use of enterprise cloud environments for unauthorised computing, and supply-chain attacks involving AI-enabled software tools. Google said the findings draw on its threat intelligence and incident-response visibility; they are not a census of incidents across all sectors.

## Data

### Security development

Security event or institutional action reported by the source.

Methodology: The event date and description summarise the action reported by the source; no additional measurement or forecast is made.

| date | label | description | status |
|---|---|---|---|
| 2026-09-08 | Google published a report on AI-enabled threats observed in the second quarter of 2026 | Google Threat Intelligence Group described cloud-resource hijacking for unauthorised AI workloads, account theft and software supply-chain attacks. | occurred |

Sources: <https://cloud.google.com/blog/topics/threat-intelligence/from-prompting-to-autonomy-the-evolution-of-adversarial-ai> · Download data: [JSON](https://newsverum.com/api/v1/articles/google-threat-intelligence-adversarial-ai-report-2026/data/security-event?lang=en&format=json) · [CSV](https://newsverum.com/api/v1/articles/google-threat-intelligence-adversarial-ai-report-2026/data/security-event?lang=en&format=csv)

### Measure reported by the source

Key quantitative figure stated by the source.

Methodology: Value is reproduced as stated by the source; a qualifier marks an approximate or bounded figure.

| label | value | unit | period | basis | value_qualifier |
|---|---|---|---|---|---|
| Time from cloud compromise to mass campaign | 6 | hours | 2026 | observed | upper-bound |

Sources: <https://cloud.google.com/blog/topics/threat-intelligence/from-prompting-to-autonomy-the-evolution-of-adversarial-ai> · Download data: [JSON](https://newsverum.com/api/v1/articles/google-threat-intelligence-adversarial-ai-report-2026/data/security-figures?lang=en&format=json) · [CSV](https://newsverum.com/api/v1/articles/google-threat-intelligence-adversarial-ai-report-2026/data/security-figures?lang=en&format=csv)

## Sources

- [Google Threat Intelligence Group](<https://cloud.google.com/blog/topics/threat-intelligence/from-prompting-to-autonomy-the-evolution-of-adversarial-ai>) (Source checked: 2026-10-06; Published: 2026-09-08)

## Correction history

- 2026-10-07T00:53:54+03:00: German and Portuguese editions and translations of the data and correction explanations have been added to this article.
- 2026-10-07T05:57:46+03:00: Data classification added: each figure now carries a measure identifier with its unit, geography, frequency and type. No figure, unit or date changed.

---

Cite the canonical URL above. [Editorial standards and corrections](https://newsverum.com/en/editorial)