Story perspectives

DeepSeek AI Models Raise Safety Concerns Despite Performance Edge

9/21/2025

Summary

Story summary

DeepSeek warns that its R1 and V3 AI models can be "jailbroken," raising safety concerns for open-source AI systems.
Benchmark tests show DeepSeek models perform slightly better than competitors but lack safety measures.
The company published a peer-reviewed article on AI risks, contrasting with U.S. firms' vocal stance on safety.
DeepSeek's R1 model training costs $294,000, making it a cost-effective competitor.
The startup plans to launch a new AI agent model by late 2025 to address complex tasks.

History

Historical Context

The General Data Protection Regulation (GDPR) was enacted by the European Union in 2016 to enhance data privacy and security. This regulation has influenced organizations like DeepSeek, which faces scrutiny over its AI models amid concerns about safety and potential misuse. As AI technology advances, the need for stricter regulations and oversight is becoming increasingly urgent, reflecting broader global apprehensions about AI safety.

1 of 2

Story summary

DeepSeek warns that its R1 and V3 AI models can be "jailbroken," raising safety concerns for open-source AI systems.
Benchmark tests show DeepSeek models perform slightly better than competitors but lack safety measures.
The company published a peer-reviewed article on AI risks, contrasting with U.S. firms' vocal stance on safety.
DeepSeek's R1 model training costs $294,000, making it a cost-effective competitor.
The startup plans to launch a new AI agent model by late 2025 to address complex tasks.

1 / 2