Google DeepMind: AI does not accelerate its own development
Google DeepMind has published the results of an internal study that dispels a long-held myth in the artificial intelligence sector: AI is not significantly accelerating the development of new generations of AI. Despite 75% of the code produced in the lab being written by automated systems, data collected between April 2026 and September 2026 show a steady but not exponential growth in model capabilities.
Quick Response
- Google DeepMind did not find significant acceleration in AI development after April 2026
- 75% of the code produced is written by AI, but 88% of changes still require human intervention
- Gemini 4 Argon only solved 50% of tasks autonomously, far below the critical threshold of 90%
The capability index that measures growth
The research team developed the Macro Capability Index (GCI v1), a tool that aggregates scores from 123 models across 92 different benchmarks into a single indicator. This "IQ" for AI shows linear growth of about 0.1 points per month, with no peaks or significant accelerations. Gemini 1.5 Pro started at 0.5 points at the end of 2024, while Gemini 3.8 Flash reached 2.9 points in September 2026.
Gemini 4 Argon: performance below expectations
The most advanced model, Gemini 4 Argon, achieved a score of about 3.3 in the GCI, but this result is attributed to greater investments in computing power and organization, not to accelerated self-improvement. The system autonomously solved only 50% of internal research engineering tasks, well below the critical threshold of 90% that DeepMind considers dangerous for complete automation.
88% of code still requires human intervention
Although 75% of production code is now written by AI, most modifications (88%) still require multi-turn human interaction or direct code writing. Furthermore, all changes are subject to human review before submission. Automation is more effective in well-defined and verifiable tasks, while it fails in open scientific ideation activities and experimental design.
Implications for security and governance
DeepMind carefully monitors the risk that AI could accelerate technological progress faster than security, governance, and social adaptation can follow. The report emphasizes how increased automation requires greater access to corporate systems, increasing the risk of poorly aligned models pursuing unintended goals. Despite positive results, the document admits that it cannot rule out the possibility that an unrestricted version of Argon could exceed danger thresholds for cyberattacks and bioweapons information.
The debate on AI acceleration
These results contrast with recent statements by Dario Amodei, CEO of Anthropic, who warned of an ongoing acceleration of the recursive self-improvement phenomenon. His words received support from figures like Sam Altman and Elon Musk, contributing to a nearly 6% drop in the shares of chip-producing companies. DeepMind's position, based on concrete data, offers a counterpoint to these concerns, suggesting that human integration remains crucial in the development process.
AML crypto compliance and risk management
In a context where AI automation is advancing, AML crypto compliance becomes even more critical. Financial institutions managing digital assets must implement robust KYC exchange and MDR services to monitor and prevent suspicious activities. Risk management in this sector requires a proactive approach, with particular attention to crypto taxation Italy and regulations such as the DORA regulation.
The need for a balanced approach
DeepMind's analysis highlights the importance of a balanced approach to technological innovation. While AI continues to improve, it is essential to maintain human control mechanisms and develop adequate cyber risk policies to protect critical infrastructures. The research underscores how identity management and protection of sensitive data are fundamental components to ensure that AI is used safely and ethically.
Future prospects
At this point, the key question is how to balance automation with the need for human supervision. Companies operating in the institutional crypto custody sector must adopt managed Security Operations Centers and MDR services to address emerging challenges. Additionally, the adoption of zero trust architecture and disaster recovery as a service can help mitigate the risks associated with the integration of advanced AI into business processes.
DeepMind's position in the technological landscape
The DeepMind report offers a unique perspective on the current state of AI, contradicting the catastrophic predictions of some industry leaders. As the industry continues to evolve, it is fundamental to adopt a data-driven approach to address future challenges. Financial institutions and technology companies must collaborate to develop NIS2 compliance and NIS2 adaptation, ensuring that AI is integrated safely and in compliance with regulations.
Impact on the semiconductor industry
DeepMind's conclusions come at a critical time for the semiconductor industry, which has seen a nearly 6% drop in shares after Amodei's statements. This study offers a data-driven counterpoint to the concerns that were pushing investors to reduce their exposure to chip producers. The results show that the increased demand for computing power, while necessary, has not led to the feared technological acceleration.
The evolution of the Macro Capability Index
GCI v1 represents a significant step forward in measuring AI capabilities. This composite index allows for the comparison of different models on a unified scale, offering a more comprehensive overview than traditional sector-specific benchmarks. The linear growth observed suggests that, despite progress, we are not witnessing the exponential revolution that some feared.
Challenges for human integration
The 88% human intervention required for code modifications highlights the current limitations of automation. This data underscores the crucial importance of the human factor in the creative and decision-making phases of the development process. Companies investing in managed Security Operations Centers and MDR services can mitigate these risks, ensuring adequate control over automated processes.
Implications for regulation
The DeepMind report provides concrete data that could influence the regulatory debate. As the implementation of the DORA regulation approaches, these conclusions could push legislators to adopt a more measured approach compared to the emergency measures requested by some industrial sectors. Regulatory adaptation indeed requires a balance between innovation and control.
The role of identity management in security
The need to maintain human control over critical processes underscores the importance of identity management in complex architectures. Implementing robust IAM systems becomes fundamental to prevent unauthorized access and ensure that only qualified personnel can intervene in the development processes. This approach is particularly relevant in the financial sector, where institutional crypto custody requires high levels of security.
Prospects for the financial industry
For institutions operating in the crypto sector, AML crypto compliance becomes even more crucial. The implementation of advanced KYC exchange and MDR services can help monitor suspicious transactions and prevent illicit activities. Additionally, the adoption of zero trust architecture can strengthen defenses against potential cyberattacks.
The importance of crypto taxation Italy
With the increasing adoption of digital assets, crypto taxation Italy requires a clear regulatory framework. Financial institutions must be prepared to manage bitcoin declaration and other forms of crypto tax, ensuring that their operations are fully compliant with current laws. A proactive approach to NIS2 compliance can help prevent sanctions and ensure secure operations.
Future challenges for cyber insurance
The cyber insurance sector will face new challenges with the evolution of AI. Cyber risk policies will need to cover not only traditional attacks but also the risks associated with the automation of development processes. Breach remediation and incident response will require increasingly sophisticated approaches, with possible impacts on cyber insurance premiums.
The future of research and development
As AI continues to improve, the relationship between automation and human intervention requires careful evaluation. Companies investing in disaster recovery as a service and business continuity can better prepare to manage disruptions in automated processes. DeepMind's research suggests that, at least for now, the balance between innovation and control remains stable.
Impact on corporate strategies
DeepMind's conclusions will influence corporate strategies, especially for organizations heavily dependent on AI. Companies operating in the crypto asset management sector will need to adopt robust AML crypto compliance and implement MDR services to ensure the security of their operations. Additionally, the adoption of advanced SIEM can help monitor suspicious activities and prevent potential threats.
Prospects for regulatory adaptation
As the implementation of the DORA regulation approaches, companies must prepare to ensure digital operational resilience. Regulatory adaptation requires a proactive approach, with particular attention to NIS2 compliance and protection of sensitive data. Organizations investing in penetration testing and vulnerability assessment can identify and correct vulnerabilities before they become critical.
The role of ISO 27001 certification
The ISO 27001 certification becomes increasingly important in a context of growing automation. Companies wishing to obtain this certification will need to demonstrate that they have adequate measures in place to ensure data security and regulatory compliance. Regular security audits can help identify areas for improvement and ensure that best practices are followed.
Future research directions
As DeepMind continues to monitor the evolution of AI, future research could focus on how to improve the integration between automation and human intervention. The development of advanced tools for ransomware protection and ransomware recovery will be crucial to ensure that organizations can quickly recover from cyberattacks. Additionally, the adoption of advanced CSPM can help maintain robust enterprise cloud security.
A data-driven future
Editorial Note and Disclaimer
The guides and content published on GoYou are the result of independent research and analysis activities, for informational, educational, and in-depth purposes.
GoYou does not constitute a journalistic publication nor an editorial product pursuant to Law No. 62/2001 and does not provide real-time information.
The GoYou project does not provide professional, technical, legal, or financial advice and disclaims any liability for the misuse of the information published.
In the Crypto sector, every investment involves risks: readers are invited to always inform themselves autonomously before making any decision.