Tim Clancy
@viewfromarchstreet.com
Arch Street=making connections in science, cybersecurity & innovation. former NSF, House Science Committee. environment, upstate NY, science & tech policy everywhere. Jazz. Occasional Boston Celtics musings
Alexandria, Virginia
viewfromarchstreet.com
Alexandria, Virginia
viewfromarchstreet.com
Reposted by Tim Clancy
Rest in power Duncan Sheik youtu.be/LX3vsw2F_DA?...
Duncan Sheik - She Runs Away (Official Video)
YouTube video by Atlantic Records
youtu.be
September 19, 2026 at 1:44 AM
Rest in power Duncan Sheik youtu.be/LX3vsw2F_DA?...
Reposted by Tim Clancy
For those interested in this topic see: www.nist.gov/artificial-i...
AI Standards
On July 29, 2026, NIST released an initia
www.nist.gov
September 18, 2026 at 10:22 PM
For those interested in this topic see: www.nist.gov/artificial-i...
Reposted by Tim Clancy
Joe Romm gave one of the most memorable talks I’ve ever heard at Columbia many years ago.
Among other things, he said, “Sustainable development is like teenage sex. Everybody says they're doing it, very few people actually are, and those who are doing it are doing it badly.”
RIP legend.
Among other things, he said, “Sustainable development is like teenage sex. Everybody says they're doing it, very few people actually are, and those who are doing it are doing it badly.”
RIP legend.
It is with tremendous sadness and sense of loss that we regretfully share the news that @penncssm.bsky.social senior research fellow, and my close personal friend Joseph Romm has passed away unexpectedly. He will be sorely missed here.
web.sas.upenn.edu/pcssm/news/r...
web.sas.upenn.edu/pcssm/news/r...
Remembering Dr. Joe Romm
It is with great sadness that the Penn Center for Science, Sustainability and the Media shares news of the passing of PCSSM Senior Research Fellow Dr. Joe Romm. Please find below an obituary prepared ...
web.sas.upenn.edu
September 18, 2026 at 8:51 PM
Joe Romm gave one of the most memorable talks I’ve ever heard at Columbia many years ago.
Among other things, he said, “Sustainable development is like teenage sex. Everybody says they're doing it, very few people actually are, and those who are doing it are doing it badly.”
RIP legend.
Among other things, he said, “Sustainable development is like teenage sex. Everybody says they're doing it, very few people actually are, and those who are doing it are doing it badly.”
RIP legend.
The road to AI safety runs through cybersecurity.
September 18, 2026 at 6:26 PM
The road to AI safety runs through cybersecurity.
They're not "hallucinations" they're probabilistic results given when the system runs out of data or has poor data. The system is performing as intended. Even at 90+% accuracy, that means *a lot* of answers are wrong.
September 18, 2026 at 5:52 PM
They're not "hallucinations" they're probabilistic results given when the system runs out of data or has poor data. The system is performing as intended. Even at 90+% accuracy, that means *a lot* of answers are wrong.
Loudon County incoming
* VIRGINIA GOVERNOR SAYS STATE IS BANNING NON-DISCLOSURE AGREEMENTS FOR DATA CENTERS, REQUIRING STRICTER PERMITTING AND CLEAN ENERGY MANDATES FOR THE SERVER WAREHOUSES
@reuters.com
@reuters.com
September 18, 2026 at 4:09 PM
Loudon County incoming
How do humans reason over rules?
We have theories, ideas but that's about it. No machine can reason over rules the way humans do because we don't know how humans do it.
For that, we need science not technology.
We have theories, ideas but that's about it. No machine can reason over rules the way humans do because we don't know how humans do it.
For that, we need science not technology.
September 18, 2026 at 5:01 AM
How do humans reason over rules?
We have theories, ideas but that's about it. No machine can reason over rules the way humans do because we don't know how humans do it.
For that, we need science not technology.
We have theories, ideas but that's about it. No machine can reason over rules the way humans do because we don't know how humans do it.
For that, we need science not technology.
And thus in part why humans use a variety of reasoning techniques such as heuristics.
And finally a more colloquial way to say it:
Stats models can only be used to accurately predict things that are relatively well-understood — not the far future! — because “unknown unknowns” can’t be estimated from reasonably obtainable data.
Stats models can only be used to accurately predict things that are relatively well-understood — not the far future! — because “unknown unknowns” can’t be estimated from reasonably obtainable data.
September 18, 2026 at 4:44 AM
And thus in part why humans use a variety of reasoning techniques such as heuristics.
A lot of OTA hagiography takes place on this free website. As a former S&T Hill staffer i can say two things were true —
1. OTA did excellent work and produced comprehensive analysis.
2. Most OTA studies were late and thus useless for their purpose of informing Congress in a timely manner.
1. OTA did excellent work and produced comprehensive analysis.
2. Most OTA studies were late and thus useless for their purpose of informing Congress in a timely manner.
September 17, 2026 at 11:43 AM
A lot of OTA hagiography takes place on this free website. As a former S&T Hill staffer i can say two things were true —
1. OTA did excellent work and produced comprehensive analysis.
2. Most OTA studies were late and thus useless for their purpose of informing Congress in a timely manner.
1. OTA did excellent work and produced comprehensive analysis.
2. Most OTA studies were late and thus useless for their purpose of informing Congress in a timely manner.
Because security audits are a thing.
Cybersecurity standards are a thing.
Disclosure requirements are a thing.
Cybersecurity standards are a thing.
Disclosure requirements are a thing.
AI CEOs are in awe that their models can, given few constraints, hack autonomously. They're also wracked with a vague fear that their products will soon hack the entire internet. I reported on how it's very curious that they seem utterly disinterested in working with cybersecurity experts on this.
Cybersecurity experts say AI giants are shutting them out of safety plans
Cybersecurity experts told NBC News they were concerned that fundamental issues of cybersecurity weren’t being addressed.
www.nbcnews.com
September 17, 2026 at 11:22 AM
Because security audits are a thing.
Cybersecurity standards are a thing.
Disclosure requirements are a thing.
Cybersecurity standards are a thing.
Disclosure requirements are a thing.
Reposted by Tim Clancy
Banning "superintelligent" 🤢 AI plays right into industry rhetoric and keeps the focus off immediate, real world harms such as privacy, bias, and labor. Literally the same name of the silly Bostrom book everyone was reading 12 years ago.
September 16, 2026 at 9:59 PM
Banning "superintelligent" 🤢 AI plays right into industry rhetoric and keeps the focus off immediate, real world harms such as privacy, bias, and labor. Literally the same name of the silly Bostrom book everyone was reading 12 years ago.
It’s a general rule in critical infrastructure that ‘squirrel’ is a much more likely cause of failure than super sophisticated cyber attack. But corporate publicity stunt is gaining momentum.
I doubt LLMs will become a digital genie or world-ending demon, but I worry one of these companies is inadvertently going to do a big cyber attack, in a publicity stunt that does more than they expected perhaps.
September 16, 2026 at 2:25 AM
It’s a general rule in critical infrastructure that ‘squirrel’ is a much more likely cause of failure than super sophisticated cyber attack. But corporate publicity stunt is gaining momentum.
Maybe the Three R's weren't so bad after all.
The Dean of Harvard College has been on a PR tour defending the institution's approach to AI and writing, but student testimony from inside the institution suggests that the school is retreating from human exchange as central to the learning experience. www.thecrimson.com/article/2026...
I’m a Writing Tutor at Harvard. The College is Failing Us. | The Harvard Crimson
The Writing Center’s dismantling evokes a broader pattern of institutions treating human learning and connection as costs to cut, not values to protect.
www.thecrimson.com
September 15, 2026 at 11:41 PM
Maybe the Three R's weren't so bad after all.
As I mentioned
This effectively ends NSF as a scientific discipline and merit-based peer review organization Whatever comes next will not be what has served the nation well for the past 75 years. www.science.org/content/arti...
Exclusive: NSF faces radical shake-up as officials abolish its 37 divisions
Changes seen as a response to presidential directives on what research to fund
www.science.org
September 15, 2026 at 10:59 PM
As I mentioned
Didn’t need a memo to call this over a year ago bsky.app/profile/view...
A deep dive into the memo that marks the end of how NSF has done business for 76 years www.science.org/content/arti...
NSF memo implementing Trump’s ‘golden age’ of science unsettles researchers
New document emphasizes support for White House priorities over curiosity-driven ideas
www.science.org
September 15, 2026 at 10:23 PM
Didn’t need a memo to call this over a year ago bsky.app/profile/view...
Reposted by Tim Clancy
I explain why a "duty of care" that makes sense in other industries completely fails in a communications and speech tool.
Australia is pushing a new "duty of care" policy to hold internet companies responsible for harms that befall their users. But applying such rules to speech works very differently than for most products.
@masnick.com and @benwhitelaw.bsky.social discuss on the latest episode of Ctrl-Alt-Speech.
@masnick.com and @benwhitelaw.bsky.social discuss on the latest episode of Ctrl-Alt-Speech.
September 13, 2026 at 4:39 PM
I explain why a "duty of care" that makes sense in other industries completely fails in a communications and speech tool.
Reposted by Tim Clancy
The paradox of semiconductor manufacturing is that foundries operate more efficiently in arid climates but require *a lot* of water even with high levels of recycling.
"Arizona’s lifeline for chip manufacturing is drying up" www.theverge.com/science/9935...
September 12, 2026 at 8:50 PM
The paradox of semiconductor manufacturing is that foundries operate more efficiently in arid climates but require *a lot* of water even with high levels of recycling.
Reposted by Tim Clancy
"Arizona’s lifeline for chip manufacturing is drying up" www.theverge.com/science/9935...
September 12, 2026 at 6:16 PM
"Arizona’s lifeline for chip manufacturing is drying up" www.theverge.com/science/9935...
The U.S. National Science Foundation and the Department of Commerce have scrubbed references to the dangers of political interference in science from official agency policy documents.
www.aip.org/fyi/science-...
www.aip.org/fyi/science-...
Science Agencies Pull ‘Political Interference’ Protections From Scientific Integrity Policies
NSF and the Department of Commerce have scrubbed references to the dangers of political interference in science from official agency policy documents.
www.aip.org
September 11, 2026 at 8:35 PM
The U.S. National Science Foundation and the Department of Commerce have scrubbed references to the dangers of political interference in science from official agency policy documents.
www.aip.org/fyi/science-...
www.aip.org/fyi/science-...
Reposted by Tim Clancy
My Talk at DEF CON
Last month, I gave a talk at DEF CON on AI hacking: what happens when AIs become hackers....
https://www.schneier.com/blog/archives/2026/09/my-talk-at-def-con.html
Last month, I gave a talk at DEF CON on AI hacking: what happens when AIs become hackers....
https://www.schneier.com/blog/archives/2026/09/my-talk-at-def-con.html
My Talk at DEF CON
Last month, I gave a talk at DEF CON on AI hacking: what happens when AIs become hackers. It's a combination of the potentialities I raised in my 2022 book A Hacker's Mind and the lessons we're learning from current AI models engaging in hacking behavior. I'm really proud of the talk, and the fact that it gained over 100K views on YouTube in just a few days. Also online is an interview with me in the AI Village.
www.schneier.com
September 11, 2026 at 6:07 PM
My Talk at DEF CON
Last month, I gave a talk at DEF CON on AI hacking: what happens when AIs become hackers....
https://www.schneier.com/blog/archives/2026/09/my-talk-at-def-con.html
Last month, I gave a talk at DEF CON on AI hacking: what happens when AIs become hackers....
https://www.schneier.com/blog/archives/2026/09/my-talk-at-def-con.html
In economic equations you will find notations for ‘all contracts complete’. What’s that you say? A giant fudge that stipulates all future states of the world are known to the contracting parties. Completely devoid of reality but necessary for the underlying math in the models to add up.
Economists call these equations "models," and they are famously impenetrable...unless you've had extensive training in economics.
September 11, 2026 at 4:46 PM
In economic equations you will find notations for ‘all contracts complete’. What’s that you say? A giant fudge that stipulates all future states of the world are known to the contracting parties. Completely devoid of reality but necessary for the underlying math in the models to add up.
AI can't function properly without a baseline of knowledge. That baseline is generated, quantified, and validated through the scientific method. Similarly, science can benefit from the use of AI but only if that AI trustworthy, validated by science.
Scoop: This afternoon NSF will formally announce its four 'Frontier Initiatives', and a suite of other changes in a public letter.
These Frontier Initiatives (left) are identical to OSTP's 'Grand Research Challenges' that we have been reporting on (right).
www.nature.com/articles/d41...
These Frontier Initiatives (left) are identical to OSTP's 'Grand Research Challenges' that we have been reporting on (right).
www.nature.com/articles/d41...
September 10, 2026 at 7:07 PM
AI can't function properly without a baseline of knowledge. That baseline is generated, quantified, and validated through the scientific method. Similarly, science can benefit from the use of AI but only if that AI trustworthy, validated by science.