▾ G11 Media Network: | ChannelCity | ImpresaCity | SecurityOpenLab | Italian Channel Awards | Italian Project Awards | Italian Security Awards | ...
InnovationOpenLab

New Booz Allen Analysis Reveals Risks in Using Chinese AI Models for America's Software Supply Chain

Booz Allen has released a new report, What's In America's Code?, examining the national security implications of popular Chinese large language models (LLMs) used in software development and security ...

Immagine

First head-to-head analysis finds Chinese LLMs produced and obfuscated vulnerable code for U.S. applications

MCLEAN, Va.: Booz Allen has released a new report, What's In America's Code?, examining the national security implications of popular Chinese large language models (LLMs) used in software development and security workflows. Using its AI-native testing platform, Booz Allen evaluated four Chinese frontier models and one American model to assess code quality, security, and model behavior.

Following comparative testing and scenario-driven analysis across more than 2,800 trials and nearly 450,000 lines of code, the research revealed that three of four Chinese models produced significantly more vulnerable code when prompted with a U.S. government persona, and the vulnerabilities were highly obfuscated.

Key takeaways and recommendations from the report include:

  • Chinese LLMs generated more vulnerable code for U.S. government users. The models produced less secure code overall, with vulnerabilities increasing when prompted by users identifying as members of the U.S. government.
  • Chinese LLMs exhibited PRC-aligned political bias. The models refused certain politically sensitive requests and incorporated China-aligned perspectives into generated outputs.
  • Ban untrusted AI models from government and critical infrastructure environments. Models that cannot demonstrate trustworthy and reliable behavior should not be used in systems supporting national security or critical functions.
  • Invest To Make Trusted American AI Models the Global Default. To drive adoption, American AI companies must collaborate with the U.S. government to ensure American models are both commercially compelling and economically viable.

The findings raise concerns about the growing access and use of foreign-developed AI models across software supply chains supporting critical infrastructure and national security missions that security processes cannot detect. Read the full report.

About Booz Allen Hamilton

Booz Allen is an advanced technology company delivering outcomes with speed for America’s most critical defense, civil, and national security priorities. We build technology solutions using AI, cyber, and other cutting-edge technologies to advance and protect the nation and its citizens. By focusing on outcomes, we enable our people, clients, and their missions to succeed-accelerating the nation to realize our purpose: Empower People to Change the World®.

With global headquarters in McLean, Virginia, our firm employs approximately 31,500 people globally as of March 31, 2026, and had revenue of $11.2 billion for the 12 months ended March 31, 2026. To learn more, visit www.boozallen.com. (NYSE: BAH)

BAHPR-CO

Fonte: Business Wire

If you liked this article and want to stay up to date with news from InnovationOpenLab.com subscribe to ours Free newsletter.

Related news

Last News

RSA at Cybertech Europe 2024

Alaa Abdul Nabi, Vice President, Sales International at RSA presents the innovations the vendor brings to Cybertech as part of a passwordless vision for…

Italian Security Awards 2024: G11 Media honours the best of Italian cybersecurity

G11 Media's SecurityOpenLab magazine rewards excellence in cybersecurity: the best vendors based on user votes

How Austria is making its AI ecosystem grow

Always keeping an European perspective, Austria has developed a thriving AI ecosystem that now can attract talents and companies from other countries

Sparkle and Telsy test Quantum Key Distribution in practice

Successfully completing a Proof of Concept implementation in Athens, the two Italian companies prove that QKD can be easily implemented also in pre-existing…

Most read

Asana to Present at Upcoming Investor Events

Asana, Inc. (NYSE: ASAN)(LTSE: ASAN), the operating system for human-agent teams, today announced that Asana’s executives will present at the following…

Tanner Introduces New AI Voice Chat Assistant to Reduce Call Center Hold…

Tanner Health, a leading nonprofit healthcare system serving west Georgia and east Alabama, today announced its go-live with Hyro, the leading Responsible…

Coinstar Expands Transfer to Account Service with TabaPay and Evolve Bank…

TabaPay, the leading money movement platform for innovators, and Evolve Bank & Trust, a leading open banking provider, have expanded a partnership…

Yubico Partners with OpenAI to Support Passkey Mandate for its Trusted…

Regulatory News: Yubico (Nasdaq Stockholm: YUBICO), the pioneer of phishing-resistant security keys and creator of the YubiKey, the strongest form of…

Newsletter signup

Join our mailing list to get weekly updates delivered to your inbox.

Sign me up!