Jen

Empathy and curiosity are underrated. In-House Tech Privacy AI | Duke U. #lawtwitter [email protected]

[Jen @Title9Jen]

For those asking why #macarthurpark

[Jen @Title9Jen]

@AGIGuardian Certain topics seem to
trigger guardrails by default. Eg regulatory analysis gets hijacked if there’s overlap w a hot button issue. No disclosure just lack of reasoning/context. Responses included “censorship” which was both irrelevant and a tween level Reddit understanding.

[Taniel @Taniel]

LOS ANGELES UPDATE: An excellent update for Nithya Raman.

She wins today's batch over Spencer Pratt by almost 21%, which is over what she needs.

Pratt's lead over Raman (for the 2nd runoff spot, behind Mayor Bass) goes down from 5.9% to 3.4%.

[Jen @Title9Jen]

@trq212 @MoritzW42 If true and this was customer data leakage, disturbing Claude views this as an outage not a bug. Very different remedies.

[Lukasz Olejnik @lukOlejnik]

The reason: Meta AI Chat bot did not implement zero trust framework? Nice summary by Anthropic "An agent permitted to read customer records, summarize information, and draft responses has clear boundaries. An agent with vague permission to "help with customer service" does not" https://t.co/DuzpksMx8P

[Jen @Title9Jen]

@killix @cyrusjohnson Would love for this to become the standard.

[Taniel @Taniel]

I'm already seeing a lot of analyses of the Los Angeles early numbers are though those are close to final. Again, analyze away, but do take recent history into account for hints of how the count will still evolve!

[Luiza Jarovsky, PhD @LuizaJarovsky]

🚨 BREAKING: President Trump has just signed an Executive Order on AI that implements a VOLUNTARY framework for AI developers to engage with the government before releasing "covered frontier models."

*Important:

Contrary to what many media outlets have written in the past few hours, the White House is not seeking a mandatory registration system or some sort of vetting scheme before models can be launched.

This is NOT the type of oversight being imposed, and this is explicitly clarified in the Executive Order itself (see the section I highlighted in blue below).

The focus of the EO is to help protect the United States' critical infrastructure, cyber defense capabilities, and national security (mainly against external attacks).

The trigger for this EO was likely Mythos, the model developed by Anthropic, which it voluntarily shared with the U.S. government for its cyberdefense-related capabilities but decided not to release to the public.

The White House used that as a blueprint for future "covered frontier models."

Again, it's not a registration system: it's voluntary.

-

>To stay up to date, subscribe to my AI governance newsletter (link below).

[Jen @Title9Jen]

@aamir1rasheed I love this in theory. But seems like encryption in use would be a violation of most frontier model provider AUPs. And/or defeat the purpose when the model need to learn from use (even in your own instance for your own use cases)? Would love to be wrong…

[California Wildfire Tracking @CaliFireTracker]

Over a year has passed since the infamous #PalisadesFire. While some time has passed, the Pacific Palisades Community is still well within the rebuilding phase. Using the @OroraTech platform's Fire Spread, Fire Intensity Simulation, and distance-to-assets features, officials could get a solid idea of where the fire may be headed and support all aspects of fire management in making informed decisions. Check out the Palisades Fire visualization here 👇https://t.co/rw20dhJdjh

[Mindset Machine  @mindsetmachine]

Be private.
Accept the loneliness.
Fix your life in silence.
No one is coming to save you.

[Arvind Jain @jainarvind]

AI architecture discipline is AI cost discipline.

Standardizing on a single AI platform can feel appealing because it seems simpler. Over time, though, it often becomes more expensive, more constraining, and harder to govern.

To control AI costs:
- Use the right model for the right job
- Invest in a horizontal context layer
- Centralize governance instead of pushing it to the edge

[Liz Hoffman @lizrhoffman]

Reminder that the ability to confidentially file for an IPO was a 2012 rule change meant to ease small companies (revenue < $1b) to the markets. It was later expanded but what are we even doing here

[Paul Sherman @PaulMSherman]

That’s funny.

[Jen @Title9Jen]

@MichaelMartocci We’re looking at @gleanwork for this. Has the permission based governance along with corp knowledge map and agnostic LLM
search/agentic capability. No affiliation at vetting stage but public co enterprise avail looking good.

[Karthik Hariharan @hkarthik]

This is probably the best route, especially for companies with a lot of nontechnical users.

Handing everyone Claude Code and asking them to go wild is going to continue to lead to mixed results.

Purposely built shared agents that can be optimized for common company workflows, and centrally managing the token usage and optimal LLM selection will be the path forward.

Power users will use direct harnesses. But everyone else needs a cleaner, more paved path.

[Yavnik Sharma @paired_electron]

is it just me or claude has lower 5h limit today? based on ccusage, today's 5h limit is only 40% of yesterday. that's less than half

[Jen @Title9Jen]

@EricBuess Claude premium 4.8 is unusable with new rate limits/tokens. One 8 page doc and two responses and hit a limit. No notice. Been using daily for 2.5 years. Same task last week wouldn’t have hit 5% usage with 4.7. Its unusable. Something has materially changed.

[Daniel Barber @gaijindan]

64% of AI-powered vendors aren't disclosing AI sub-processors in their legal documentation.

That was the most surprising finding from the Privacy & AI Trends Report, released today.

Thanks to @MichaelFNunez @VentureBeat for covering what this means for privacy, security, legal, and AI governance teams: https://t.co/S7G5yATCRu

If AI governance is on your roadmap this year, repost so the leaders in your network see it

[Taffy Brodesser-Akner @taffyakner]

This @WIRED package is the most hilarious and potent AI explainer — done as "mandatory AI training — and I cannot get over it. https://t.co/OxNmY3IQbi