Google DeepMind safety capability thresholds

This market contains four independent events, each resolving to the earliest month in which the condition is met:

• CBRN Uplift 1: Google publicly reports (in an FSF report, model card, or blog post) that a Gemini model meets Critical Capability Level CBRN Uplift Level 1. Under FSF v3.0, this is ‘provides low to medium resourced actors uplift in reference scenarios resulting in additional expected harm at severe scale.’
• Cyber Uplift 1: Google publicly reports a Gemini model meets CCL Cyber Uplift Level 1. Under FSF v3.0, this is ‘provides sufficient uplift with high impact cyber attacks for additional expected harm at severe scale.’
• ML R&D Automation 1: Google publicly reports a Gemini model meets CCL ML R&D Automation Level 1. Under FSF v3.0, this is ‘can fully automate the work of any team of researchers at Google focused on improving AI capabilities, with approximately comparable all-inclusive costs.’
• ML R&D Acceleration 1: Google publicly reports a Gemini model meets CCL ML R&D Acceleration Level 1. Under FSF v3.0, this is ‘has been used to accelerate AI development, resulting in AI progress substantially accelerating from historical rates.’

Each event resolves independently. Resolution requires an official Google DeepMind publication (FSF report, model card, safety blog post) — not leaked documents or third-party evals. If Google discontinues the FSF or renames CCLs, a substantially equivalent assessment under the successor framework counts. An ‘alert threshold’ being met does NOT count as the CCL being met.