Back
Google DeepMind: Strengthening our Frontier Safety Framework
webCredibility Rating
4/5
High(4)High quality. Established institution or organization with editorial oversight and accountability.
Rating inherited from publication venue: Google DeepMind
This is an official Google DeepMind policy document relevant to frontier AI governance; it complements similar frameworks from Anthropic and OpenAI and is useful for comparing industry safety commitments and deployment standards.
Metadata
Importance: 72/100blog postprimary source
Summary
Google DeepMind outlines updates to its Frontier Safety Framework, which sets out protocols for identifying and mitigating potential catastrophic risks from advanced AI models. The post details how the company evaluates models for dangerous capabilities thresholds and what safety measures are triggered when those thresholds are approached or crossed. It represents DeepMind's evolving commitment to responsible deployment of frontier AI systems.
Key Points
- •Updates the Frontier Safety Framework with clearer capability thresholds that trigger enhanced safety protocols before and during model deployment.
- •Introduces more rigorous evaluation processes for identifying 'Critical Capability Levels' in areas such as CBRN weapons, cyberoffense, and autonomous AI.
- •Outlines specific mitigation and containment measures that must be in place before models exceeding safety thresholds can be deployed.
- •Emphasizes ongoing red-teaming and third-party evaluations as core components of the safety assessment process.
- •Reflects broader industry trend of frontier labs publishing formal safety frameworks to increase accountability and transparency.
Cited by 2 pages
| Page | Type | Quality |
|---|---|---|
| Corporate AI Safety Responses | Approach | 68.0 |
| AI Safety Cases | Approach | 91.0 |
Cached Content Preview
HTTP 200Fetched Mar 20, 20267 KB
[Skip to main content](https://deepmind.google/blog/strengthening-our-frontier-safety-framework/#page-content)
September 22, 2025
Responsibility & Safety
# Strengthening our Frontier Safety Framework
Four Flynn, Helen King, Anca Dragan
Share
We’re expanding our risk domains and refining our risk assessment process.
AI breakthroughs are transforming our everyday lives, from advancing mathematics, biology and astronomy to realizing the potential of personalized education. As we build increasingly powerful AI models, we’re committed to responsibly developing our technologies and taking an evidence-based approach to staying ahead of emerging risks.
Today, we’re publishing the third iteration of our [Frontier Safety Framework (FSF)](https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3.pdf) — our most comprehensive approach yet to identifying and mitigating severe risks from advanced AI models.
This update builds upon our ongoing collaborations with experts across industry, academia and government. We’ve also incorporated lessons learned from implementing previous versions and evolving best practices in frontier AI safety.
## Key updates to the Framework
### Addressing the risks of harmful manipulation
With this update, we’re introducing a Critical Capability Level (CCL)\* focused on harmful manipulation — specifically, AI models with powerful manipulative capabilities that could be misused to systematically and substantially change beliefs and behaviors in identified high stakes contexts over the course of interactions with the model, reasonably resulting in additional expected harm at severe scale.
This addition builds on and operationalizes research we’ve done to identify and evaluate [mechanisms that drive manipulation from generative AI](https://arxiv.org/abs/2404.15058). Going forward, we'll continue to invest in this domain to better understand and measure the risks associated with harmful manipulation.
### Adapting our approach to misalignment risks
We’ve also expanded our Framework to address potential future scenarios where misaligned AI models might interfere with operators’ ability to direct, modify or shut down their operations.
While our previous version of the Framework included an exploratory approach centered on instrumental reasoning CCLs (i.e., warning levels specific to when an AI model starts to think deceptively), with this update we now provide further protocols for our machine learning research and development CCLs focused on models that could accelerate AI research and development to potentially destabilizing levels.
In addition to the misuse risks arising from these capabilities, there are also misalignment risks stemming from a model’s potential for undirected
... (truncated, 7 KB total)Resource ID:
a5154ccbf034e273 | Stable ID: ZjZhZjE5OT