WASHINGTON, D.C. — The National Security Agency has issued an advisory warning that artificial‑intelligence tools have uncovered “critical” software vulnerabilities affecting more than 70 % of publicly‑facing websites, according to a statement released on Tuesday.

Researchers from the agency’s Cybersecurity Directorate used large‑language models from OpenAI and Google to scan code repositories for patterns that could be exploited. The effort yielded 1,274 previously unknown flaws, 312 of which the NSA classifies as “high‑impact” because they could enable remote code execution on content‑management systems that power e‑commerce and news sites.

“AI‑driven analysis is changing the speed at which we can identify systemic risk,” said Dr. Elena Ramirez, senior cyber‑risk analyst at the NSA. “Enterprises that rely on open‑source web stacks must prioritize patching these findings or face potential data breaches.” The agency recommends immediate updates for WordPress, Joomla and Drupal installations, which together host roughly 45 % of the web.

The discovery comes as large corporations accelerate AI adoption for development pipelines. A recent survey by the Enterprise Software Alliance showed that 62 % of Fortune 500 firms plan to integrate generative AI into security testing by the end of 2025, a move that could both mitigate and amplify exposure if the underlying models are not properly vetted.

Industry analysts note that the competitive landscape is shifting. While OpenAI’s GPT‑4 and Google’s Gemini are praised for their code‑analysis capabilities, smaller vendors such as Anthropic and Cohere are racing to offer “secure‑by‑design” AI suites that limit data leakage during scanning. “The market will likely consolidate around providers that can certify their models against government‑issued baselines,” said Michael Chen, a senior analyst at TechInsights.

Cyber‑security firms have already begun releasing patches and advisory bulletins. Microsoft reported that its Azure Security Center will incorporate the NSA’s findings into its automated vulnerability assessment module next month, potentially protecting millions of cloud‑hosted applications.

Key Takeaways

• AI tools uncovered 1,274 new software flaws, 312 deemed high‑impact.
• Over 70 % of websites use affected platforms; immediate patching is urged.
• Enterprise AI adoption may accelerate both remediation and new risk vectors.

Enterprises are advised to inventory their web‑application stack, apply the latest security patches, and consider AI‑enhanced testing while monitoring guidance from the NSA and industry partners.