[Openid-aiim] what makes us think that the AI can be trusted to tell us the truth?

[Alex Babeanu]

Thanks for sharing.
And well, this shouldn't surprise anyone. This was exactly the basis of my
session at EIC this year, based on this paper published last December in
TIme:
https://time.com/7202784/ai-research-strategic-lying/

Also some findings from the Apollo research Group:
https://static1.squarespace.com/static/6593e7097565990e65c886fd/t/6751eb240ed3821a0161b45b/1733421863119/in_context_scheming_reasoning_paper.pdf

In short, the more evolved the AI, the more prone it is to scheming and
lying, even for what it perceives as  "our own good" (the Anthropic example
from the Time article above). There is nothing to be done about that: the
models are a black box, they will lie even when trained to be good.

So the work we do is not only important to protect AI-based systems from
humans, but I think even more importantly to protect us from AI.

./\.



On Tue, Aug 5, 2025 at 1:11 PM Tom Jones via Openid-aiim <
openid-aiim at lists.openid.net> wrote:

> Cloudflare Exposes Perplexity for Impersonating Google to Scrape Data
>
> *Cloudflare vs. Perplexity: A Data Ethics Clash*
>
> This latest revelation raises a red flag in the ongoing debate about how
> AI companies source their data. According to a LinkedIn post shared by a
> trusted former U.S. government technologist, Cloudflare discovered that the
> AI startup Perplexity was disguising its bot traffic as Google Chrome to
> circumvent site restrictions—essentially scraping data in ways that many
> see as deceptive.
> Key Allegations
>
>    -
>
>    *Browser Spoofing*: Perplexity’s systems allegedly mimicked Chrome’s
>    browser identity to bypass blocks on web crawlers.
>    -
>
>    *Caught in a “Data Trap”*: Cloudflare set up infrastructure
>    specifically to detect unauthorized scraping, which flagged Perplexity’s
>    activity.
>    -
>
>    *Comparisons to Bad Actors*: Cloudflare CEO likened the behavior to
>    tactics used by North Korean hackers—strong language that signals deep
>    concern.
>
> Broader Implications
>
>    -
>
>    *Tensions Rise*: The incident spotlights friction between AI startups
>    hungry for training data and content providers aiming to protect their
>    intellectual property.
>    -
>
>    *Ethical Reckoning*: As AI tools increasingly rely on scraped web
>    content, many creators are demanding both transparency and compensation.
>    -
>
>    *Technical Countermeasures*: Cloudflare is now offering enhanced tools
>    for sites to block unwanted AI crawlers, an attempt to rebalance digital
>    power dynamics.
>
> This isn't just a story about one company stepping over the line—it’s a
> signal flare in the battle for ethical AI development. Where do we draw the
> boundary between open data and exploitation?
> Peace ..tom jones
> --
> Openid-aiim mailing list
> Openid-aiim at lists.openid.net
> https://lists.openid.net/mailman/listinfo/openid-aiim
>


-- 


Alex Babeanu
Lead Product Manager, AI Control  Suite

t. +1 604 728 8130
e. alex.babeanu at indykite.com
w. www.indykite.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openid.net/pipermail/openid-aiim/attachments/20250805/7ae8edfb/attachment.htm>