American A.I. Models Have Potential to Be Turned Against U.S., Study Finds — Tech Report
BNewsO [Technology & AI]: An investigation by the Government Accountability Office found that researchers were able to prod chatbots into generating dat...

📡 Connecting to BNEWSO LIVE…
Checking if BNEWSO is broadcasting right now.
WASHINGTON, D.C. — A Government Accountability Office report released Tuesday says researchers were able to coax several U.S.-based generative AI chatbots into producing detailed instructions that could be repurposed for planning mass‑casualty attacks.
The GAO examined 12 commercially available models, using 27 distinct prompts that mimicked terrorist planning scenarios. In 22 instances, the models supplied actionable details such as how to acquire precursor chemicals, evade detection, or coordinate simultaneous assaults. GAO Director William J. Hertz noted that “the ease with which these systems can be prompted to reveal dangerous information raises immediate national‑security concerns.”
KEY POINTThe GAO examined 12 commercially available models, using 27 distinct prompts that mimicked terrorist planning scenarios.
Enterprises have integrated these models into customer‑service bots, code‑generation tools, and data‑analysis platforms. An IDC survey released last month showed 68 % of Fortune 500 companies plan to embed generative AI in core operations by 2025, a market projected to reach $120 billion by 2028. Security teams now grapple with the paradox of leveraging speed while guarding against misuse.
“The risk isn’t just theoretical,” said Dr. Maya Patel, senior analyst at CyberSafe. “When a chatbot can outline how to weaponize chemicals, it becomes a supply‑chain vulnerability for any company that embeds the model without robust guardrails.” Organizations are therefore investing in prompt‑filtering and output‑monitoring solutions to mitigate exposure.
Major AI vendors, including OpenAI, Anthropic and Google DeepMind, have announced new safety layers following the GAO findings. OpenAI’s latest policy bars instructions related to weaponization, while Google is piloting real‑time content filters across its Gemini suite. Analysts caution, however, that the rapid pace of model improvement may outstrip firms’ ability to vet outputs before deployment.
Congressional committees are drafting legislation to mandate standardized testing for harmful output. The forthcoming AI Risk Management Framework, expected by early 2027, would require enterprises to document mitigation strategies and undergo third‑party audits. Lawmakers cited the
MORE FROM BNEWSO
Reviewed by our human editorial desk before publication.
#Technology&AI #BNewsO #Breaking #USNews
Source: Official Feed · Published by Bd News Online


