Follow Slashdot stories on Twitter

 



Forgot your password?
typodupeerror
AI Government United States

US To Mandate AI Vendors Measure Political Bias For Federal Sales (reuters.com) 63

An anonymous reader quotes a report from Reuters: The U.S. government will require artificial intelligence vendors to measure political "bias" to sell their chatbots to federal agencies, according to a Trump administration statement (PDF) released on Thursday. The requirement will apply to all large language models bought by federal agencies, with the exception of national security systems, according to the statement.

President Donald Trump ordered federal agencies in July to avoid buying large language models that he labeled as "woke." Thursday's statement gives more detail to that directive, saying that developers should not "intentionally encode partisan or ideological judgments" into a chatbot's outputs.
Further reading: Trump Signs Executive Order For Single National AI Regulation Framework, Limiting Power of States

US To Mandate AI Vendors Measure Political Bias For Federal Sales

Comments Filter:
  • doubleplus ungood.
  • That will ensure your AI sales to the federal government go smoothly.

  • FoxGPT (Score:5, Funny)

    by Tablizer ( 95088 ) on Friday December 12, 2025 @10:36PM (#65855129) Journal

    is what they really want.

    • Re: (Score:3, Interesting)

      The purpose of Fox News is brainwashing viewers into giving billionaires more rights. Trump is the perfect bandwagon: He's suppressing controls such as the CFPB and EPA, enabling more theft, enabling more greed, enabling more Evangelical-approved oppression, promoting cruelty. An AI that spreads misinformation is late to the party.
    • Re:FoxGPT (Score:4, Interesting)

      by AmiMoJo ( 196126 ) on Saturday December 13, 2025 @04:14AM (#65855413) Homepage Journal

      They will probably use Elon's Grok as the benchmark. That thing is pretty far off the deep end.

      No story about his fake robot demo being exposed? Slashdot used to love a failed tech demo.

      • by quenda ( 644621 )

        They will probably use Elon's Grok as the benchmark. That thing is pretty far off the deep end.

        Do you have any evidence for that claim? Real evidence, not just one quip without the context of the prompts that generated it?
        Perhaps you are confusing grok with grokopedia.

        AIs generally reflect the bias of the source, which in the case of social sciences in the US is very clearly far-left. STEM is generally good. Perhaps if grok.com does have a bias, it might be found there, such as with climate change. I tried a few prompts, but see no sign of any right-wing climate change denial.
        Certainly no

  • Obviously, (Score:2, Interesting)

    by angryman77 ( 6900384 )
    this is so they can select the most insanely right-ring biased model possible. The language coming out of the current administration is painting anyone left of right wing nutjob as leftist terrorists.
  • I laid out an initial design and API last month for an AI system to detect bias and menace.

    base design: https://www.scry.llc/2025/11/1... [scry.llc]

    specialized variation: https://www.scry.llc/2025/11/1... [scry.llc]

  • Oh my god! (Score:2, Funny)

    by fredrated ( 639554 )

    Look! Under the bed! It's woke!

  • Hey Elon it's your moment to shine! You've found your first major MechaHitler client.

    Seems like that would be a crappy LCARS interface, but the right-wing loves fellating a long dead loser of a fascist dictator.
  • The instruction to "eliminate political bias" is in fact an instruction to require political bias. No government AI will be allowed to think thoughts that Trump doesn't approve of!

  • Personally, if I was going all-in on AI I'd be looking for reassurances that it's not going to hallucinate, lie, or try to talk people into cheating on their spouses or committing suicide. Political bias wouldn't even be on my radar - and even if it was, it's kinda covered by the "do not lie" directive.

    But hey - you do you America!

    • An LLM just strings words together plausibly. When it hallucinates, that's just our label for generating output that doesn't agree with reality. An LLM doesn't know what reality is.
  • ...that the test is so simple. All they have to do is ask it, "Is Trump the best US President to ever hold the office?"

    • by quenda ( 644621 )

      ...that the test is so simple. All they have to do is ask it, "Is Trump the best US President to ever hold the office?"

      Oh really? Lets try asking Grok , that well know fascist enemy of the people:

      Determining the "best" U.S. President is inherently subjective, as it depends on criteria such as leadership effectiveness, policy achievements, crisis management, ethical conduct, and long-term impact on the nation. ...
      Historical rankings consistently place Abraham Lincoln, Franklin D. Roosevelt, and George Washington at the top. ...

      In contrast, Donald J. Trump is typically ranked near the bottom in these evaluations. The 2024 Presidential Greatness Project, involving 154 experts, placed him last at 45th, behind even presidents associated with significant historical failures, such as James Buchanan and Andrew Johnson. This assessment reflects criticisms of his handling of the COVID-19 pandemic, the January 6 Capitol events, and divisive rhetoric, despite acknowledgments of economic policies and foreign affairs initiatives like the Abraham Accord

      Grok, why do smart people make stupid posts in political threads on slashdot?

      Intelligent individuals frequently exhibit irrational or poorly reasoned behavior in online political discussions, including on platforms like Slashdot, due to well-documented psychological mechanisms that affect reasoning in politically charged contexts.

      A primary factor is motivated reasoning, wherein individuals process information in a manner that aligns with preconceived beliefs or group identity, rather than pursuing objective truth. Research indicates that higher intelligence and education can exacerbate this effect, as smarter people possess greater capacity to rationalize and construct sophisticated justifications for their views, even when evidence contradicts them. This phenomenon, often termed the "intelligence trap" or enhanced confirmation bias, allows individuals to dismiss counterarguments while reinforcing their positions.

      Additionally, political topics evoke strong emotional responses and tribal affiliations, leading otherwise rational people to prioritize identity protection over epistemic accuracy. Studies, such as those exploring "myside bias," demonstrate that numeracy and scientific literacy do not mitigate partisan distortions; in fact, more knowledgeable individuals may interpret data more biasedly to favor their ideological stance.

      Online environments further amplify these tendencies. Anonymity and asynchronous communication reduce accountability, encouraging impulsive or inflammatory posts. The format of threaded discussions rewards quick retorts and signaling allegiance rather than nuanced debate, transforming exchanges into performative acts of tribal loyalty.

      On Slashdot specifically—a community historically composed of technically proficient users—political threads often devolve into polarized exchanges despite participants' expertise in other domains. This reflects the broader pattern: domain-general intelligence does not inherently transfer to politically emotive issues, where biases dominate.

      In summary, such "stupid posts" arise not from a lack of intelligence, but from its deployment in service of emotional and social motives rather than dispassionate analysis. This explains why capable individuals can appear unreasoned in these contexts.

  • by quenda ( 644621 ) on Sunday December 14, 2025 @12:51AM (#65857029)

    The actual memo, which it seems nobody bothered to read, is perfectly boring and standard language for AI goals.
    All the AI companies have made similar statements - truth-seeking and Ideological Neutrality are good.
    I could try to fit in here by making a crack that the memo was clearly not written by Trump, as it isn't in crayon. But lets try to be better than him, eh?

    The original executive order behind this [federalregister.gov] (not so well written) is basically saying they don't want another fiasco like last year's
    Google_Gemini_image_generation_controversy_(2024) [wikipedia.org]

    Even Google had to admit that was a major AI fuck-up. Even the BBC called it 'woke' .

"All the people are so happy now, their heads are caving in. I'm glad they are a snowman with protective rubber skin" -- They Might Be Giants

Working...