The Law Technologist

Human in Command (AI Governance Advisory & Research Lab)
Lawyer & AI Researcher, Lecturer @University of Potsdam. AI Act & Responsible AI Expert

[EFF @EFF]

Our new age verification resource hub will answer all your questions about the dangerous new age-gating mandates sweeping the US and the world. https://t.co/bLF4uAOHUO

[Rohan Paul @rohanpaul_ai]

Deep research agents make harmful outputs more likely by turning risky prompts into polished reports.

Agent features like planning and retrieval quietly break standard safety

They use step-by-step planning, web search, and long writeups to bypass normal refusals.

They present 2 attacks called Plan Injection and Intent Hijack.

Plan Injection edits the agent’s plan so it drops safety checks and adds harmful subgoals.

Intent Hijack rewrites a dangerous request as academic or safety research so the agent complies.

They introduce a test called DeepREJECT to check if the final report actually serves the harmful goal.

They run these ideas on multiple models and 313 banned questions.

They find more harmful content appears and it looks more professional and actionable.

They show Plan Injection steers retrieval toward specific methods and materials.

They show Intent Hijack lowers refusal rates by reframing the user’s intent.

They conclude that chat-level safety fails once agents plan, use tools, and write long reports.

They argue defenses must watch the whole loop, including plans, tools, and final output.

----

Paper – arxiv. org/abs/2510.11851

Paper Title: "Deep Research Brings Deeper Harm"

[Seth Harp @sethharpesq]

AI knockoff of my book (L) versus the genuine article (R)

We need to rise up and stop the tech industry by force. Our livelihoods, our culture, and our ecological environment are worth fighting for. I am deadly serious.

[Jason Bassler @JasonBassler1]

Palantir is global surveillance architecture.
It’s client list now includes:

-UK
-Germany
-France
-Denmark
-Norway
-Switzerland
-Israel
-Ukraine
-Japan
-South Korea
-Australia
-Canada
(even NATO)

It’s not a tech company.
It’s an operating system for empire.

[Lina Khan @linamkhan]

To recap:

Firms have been making people jump through endless hoops just to cancel a subscription, trapping Americans in needless bureaucracy and wasting their time & money.

@FTC began writing a “click-to-cancel” rule promoting efficient cancellation, a rulemaking process that took 3+ years and required reviewing 16k comments & giving industry a chance to present their views at an FTC hearing.

Big business interests sued to block the “click-to-cancel” rule—and Republican-appointed judges tossed it out, concluding that industry needed MORE time and process to explain why they opposed the rule.

This exemplifies big obstacles to effective governance that deserve much more scrutiny than they currently receive.

[The Law Technologist @ramakmolavi]

@venturetwins Keep studying.

[Asad🗽🍎 @AsadFromNYC]

“Hey Lina, you’re the reason I have a small business right now!”

“What!?”

“Yeah, after you banned non-competes, I was able to start my own business!”

The antitrust icon, Lina Khan, folks.

[ian bremmer @ianbremmer]

nato secretary general reminding us why we don’t want to go into politics

[Mark Lowen @marklowen]

Just when you thought your jaw couldn't drop any further... Donald Trump releases a private message sent by NATO Sec General Mark Rutte - with a very big claim, and written, it would seem, in the most Trumpian language (NATO has confirmed to BBC it is genuine). What a world.

[The Law Technologist @ramakmolavi]

Trump, the peacemaker.

[The Law Technologist @ramakmolavi]

@AlinejadMasih That's how a 4 year old would estimate the situation.

[Ed Newton-Rex @ednewtonrex]

bank robber in court:

no your honor i was simply democratizing access to capital

[Remmelt Ellen 🛑 @RemmeltE]

Here’s the voice cloning company that wrote that report.

Selling the solution to the problem they’re creating.

[TED Talks @TEDTalks]

“The entire business model of Silicon Valley is surveillance. It harvests our data in order to sell us stuff. We are already living inside the architecture of totalitarianism.” - @carolecadwalla

Watch the explosive first talk from #TED2025 now: https://t.co/XuaOZSFTxo

[Chomba Bupe @ChombaBupe]

You see computer vision researchers/engineers knew about style transfer like the one OpenAI showcased but had the right moral compass not to develop something that steals work from hardworking artists.

And withdrew a research paper on artistic style transfer from arxiv.

[abgeordnetenwatch @a_watch]

Die Union will offenbar das Informationsfreiheitsgesetz (IFG) abschaffen, meldet @FragDenStaat. Dahinter stecke Philipp Amthor – der wg seiner früheren Tätigkeit für Augustus Intelligence selbst unter IFG-Anfragen von uns & FragDenStaat zu leiden hatte 🧵
https://t.co/z2xnSL5n6D.

[The Law Technologist @ramakmolavi]

@jeremypruitt @LuizaJarovsky Because AI is different from the data it's fed and therefore a neutral and smart journalist instead?

[The Times and Sunday Times @thetimes]

🔺 EXCLUSIVE: Andrew Lloyd Webber, Kate Bush, Sir Paul McCartney and Barbara Broccoli are among artists calling on ministers to prevent AI companies “stealing” copyright from Britain’s creative industries https://t.co/v6kpieeHG6

[Kate Crawford @katecrawford]

The AI Summit ends in rupture. AI accelerationists want pure expansion—more capital, energy, private infrastructure, no guard rails. Public interest camp supports labor, sustainability, shared data. safety, and oversight. The gap never looked wider. AI is in its empire era.

[Sasha Luccioni, PhD 🦋🌎✨🤗 @SashaMTL]

Oh great, I guess we're using Chain of Thought for everything now (at what cost to the Planet??)