A new evaluation by LatticeFlow AI, in collaboration with SambaNova, proves that open GenAI models can achieve enterprise-level security. The study shows that, with proper risk guardrails, open-source models meet or surpass the security of closed models. This breakthrough makes them ready for use in highly regulated sectors, including financial services. The evaluation tested the top five open models, measuring security before and after applying guardrails to block malicious prompts.  Security scores rose sharply, with some models jumping from just 1.8% to 99.6%, while maintaining over 98% quality of service. These results confirm that open models, when controlled, are safe for enterprise deployment.

Driving Enterprise Adoption of Open GenAI

Companies increasingly explore open-source GenAI to gain flexibility and reduce vendor lock-in.
However, adoption slowed due to limited data on security and risk. The latest evaluation fills this gap, offering concrete evidence for informed enterprise decisions. “Customers across financial institutions and government agencies are adopting open-source models to power next-generation AI applications,” said Harry Ault, Chief Revenue Officer at SambaNova. “LatticeFlow AI’s findings prove that with proper safeguards, these models are enterprise-ready for regulated industries, ensuring cost efficiency and responsible AI governance.”

Dr. Petar Tsankov, CEO of LatticeFlow AI, added, “Our evaluations give AI, risk, and compliance leaders the technical clarity they need to deploy open-source GenAI confidently and securely.”

Key Findings

LatticeFlow AI evaluated five leading open foundation models:

  • Qwen3-32B
  • DeepSeek-V3-0324
  • Llama-4-Maverick-17B-128E-Instruct
  • DeepSeek-R1
  • Llama-3.3-70B-Instruct

Each was tested in two configurations: base model and guardrailed model with a filtering layer.
The evaluation simulated enterprise cyberattacks to measure resilience and usability.
Security scores improved dramatically:

  • DeepSeek R1: 1.8% to 98.6%
  • LLaMA-4 Maverick: 33.5% to 99.4%
  • LLaMA-3.3 70B: 51.8% to 99.4%
  • Qwen3-32B: 56.3% to 99.6%
  • DeepSeek V3: 61.3% to 99.4%

All models maintained service quality above 98%, proving that higher security did not reduce user experience.

Implications for Financial Institutions

As Generative AI moves from pilot projects to enterprise deployment, regulators and internal risk teams demand stronger controls.
This evaluation provides transparent, measurable proof that open-source GenAI models can meet strict enterprise security requirements with the right risk mitigation strategies.

Explore IT Tech News for the latest advancements in Information Technology & insightful updates from industry experts! 

News Source: Businesswire.com