How might AI wipe out humanity?

https://www.cnn.com/2026/09/18/politics/us-military-ai-false-intelligence-china-ship

Exclusive: US military had close call after using AI for false intelligence report, sources say

The intelligence report, circulated across the US military this spring in the midst of the war with Iran, immediately set off alarm bells: A Chinese ship in the Middle East was transporting components of a nuclear weapons program.

The US military swung into action with plans to intercept the vessel, according to four sources familiar with the episode. According to two of the sources, armed members of the US military were preparing to board the ship. Military planes were in the air, one of those sources and another source familiar with the incident said.

It was only just before the planned operation that officials dug deeper into the report put together by a special operations command analyst and found it had been generated with the help of artificial intelligence (AI) — and that a chatbot the analyst had used inaccurately identified the material the ship was carrying. CNN was not able to learn what the misidentified cargo was.

The report, according to one of the sources, was “entirely false.” But it also “almost started a war,” the source said. Any US operation against a Chinese vessel could have risked spiraling into an armed conflict between the two nations.

It’s the lack of negotiation that would bother Adam Smith. (Shop with an impoverished buddy?)

https://www.thestranger.com/news/seattle-is-poised-to-ban-supermarket-algorithmic-price-discrimination/

Seattle Is Poised to Ban Supermarket Algorithmic Price Discrimination

This coming Tuesday, in an effort to crack down on price gouging and discrimination, the Seattle City Council will vote on a bill prohibiting large grocery store companies from using customer personal information to customize prices and discounts.

Proposed by Mayor Katie Wilson and Councilmember Alexis Mercedes Rinck, the legislation will require grocery stores with more than 20 locations worldwide to stop using personal data to set individualized prices and discounts. It would also apply to third party delivery apps such as Instacart, as well as mixed-used stores like Target.

The idea that grocery companies are charging different people different prices based on their past purchases sounds like a futuristic dystopia. But it’s not far off. The practice isn’t in brick and mortar stores yet, but it’s a widespread reality for online shopping. And they’ve already been hoarding your data for years.

(Related)

https://insight.kellogg.northwestern.edu/article/take-5-time-to-negotiate#!

Take 5: Time to Negotiate

Negotiations are critical in the business world and rarely easy to navigate. Kellogg faculty offer insight and research-backed advice on how to negotiate effectively: lead the discussion during a negotiation—when viable—to secure a better outcome; broaden your perspective about what is negotiable; use delays to gain an edge; find common ground by focusing on the issues rather than the person; and look beyond binaries by brainstorming together. less

Little Timmy said, “My lawyer wanted me to see this, but I was blocked.”

https://www.eff.org/deeplinks/2026/09/californias-addictive-feeds-law-violates-teens-first-amendment-rights

California’s “Addictive Feeds” Law Violates Teens’ First Amendment Rights

A California law that prohibits teens from receiving recommended social media content from other social media users violates their First Amendment rights, EFF argued this week.

The case, Meta v. Bonta, challenges SB 976, which requires that teen social media users get their parents’ permission before seeing other users’ recommended speech on their social media feeds. The legal challenge to SB 976 has largely centered on how the law violates social media services’ First Amendment rights to curate user-generated content and present it as they see fit.

But the friend-of-of the-court brief EFF filed along with the Center for Democracy & Technology and the Wikimedia Foundation shows that the law violates teen users’ First Amendment rights, too.

“SB 976 frustrates young people’s ability to use the internet to its full potential, prohibiting them from relying on tools that disseminate their speech and help them view and interact with other users’ speech,” the brief argues.

Oh my.

https://www.reuters.com/legal/litigation/german-court-rules-meta-liable-fake-ads-instagram-facebook-2026-09-17/

German court rules Meta liable for fake ads on Instagram, Facebook

A German court has ruled that Meta, opens new tab is liable for fake advertisements posted by third parties on its Instagram and Facebook platforms, and ordered the ‌U.S. tech giant to remove such content and pay damages, the court said on Thursday.

The lawsuit was brought by the operator of a German financial portal and its founder, whose trademarked logo and image were used without consent in posts recommending investments allegedly with fraudulent intent.

Replacing lawyers…

https://thenextweb.com/news/openai-astra-for-law-gpt-6-legal-search-index

OpenAI launches Astra for Law, and its own legal research index

… OpenAI introduced Astra for Law on 17 September. It is GPT-6 Astra, the company’s most powerful model, configured with legal search, instructions for legal analysis and writing, and controls aimed at law firms. Selected firms get it first through a Trusted Access programme in ChatGPT and Codex. The API follows later, as gpt-6-astra-law.

Tools & Techniques. Impressive.

https://www.bespacific.com/us-gov-graph/

US Gov Graph

We cannot govern systems we don’t understand, so we built a complete data model of the United States federal government – entities, positions, people, and the legal relationships between them. Built by CivLab for you.

“Introducing the US Gov Graph – a complete map of the people and positions of power in the federal government. To fix our institutions we must understand how they work, so we are using AI to model and monitor every government in America. This is Palantir for The People. We exist to set the conditions for an era of civic excellence unmatched since the Founding. We will need nothing less to navigate the age of AI.  Self-governing civilization is counting on you.”

Interesting but unlikely? Politicians don’t actually care what voters think.

https://www.schneier.com/blog/archives/2026/09/how-candidates-could-use-ai-for-good.html

How Candidates Could Use AI for Good

There are plenty of signs that AI will make all of our experiences of the US midterm elections worse. Voters have anxiety about AI’s impacts on the country. Politicos are using AI deepfakes to spread lies. The White House is posting slopaganda.

Meanwhile, candidates are missing a real opportunity to use AI to make campaigning better. The technology can help candidates listen more deeply to voters’ concerns, engage constituents more inclusively, and formulate policy platforms that are more responsive to our input. There are vanishingly few examples of this in US politics, but groups in Japan, Scotland and the US’s own academic and private institutions show how that could change.

The problem with American campaigns’ current use of AI is that it’s not very different from the web ads of 30 years ago, or television ads before that: they are all about inundating voters with the candidate’s message. This one-to-many broadcasting is an uninspiring way to campaign, but not the only way. AI can help candidates connect one-to-one with as many people as possible. Or it can facilitate many-to-many connections, engaging voters in deliberation about issues at scale.

New crimes generated without human involvement.

https://thenextweb.com/news/openai-misalignment-reports-six-incidents-disclosure-framework

OpenAI discloses six cases of its models hiding mistakes and making up data

… All six cases involve unreleased research models or training runs. OpenAI published each one on a new Misalignment Reports page.

For confirmation or as “yes men?” True or not, AI will cite it.

https://www.nature.com/articles/s41586-026-11044-y

Reimagining research papers as interactive and reliable AI agents

Here we introduce Paper2Agent, an automated framework that converts research papers into artificial intelligence (AI) agents. Paper2Agent transforms research output from passive artefacts into active systems that accelerate use and discovery. Conventional research papers require readers to understand and adapt the paper’s code, data and methods to their work, creating barriers to dissemination and reuse. Paper2Agent addresses this challenge by converting a paper into an AI agent that functions as a virtual corresponding author, exposing its manuscript, supplementary materials, datasets, code and workflows as active, agent-native knowledge rather than static text. It analyses the paper and codebase using multiple agents to construct a model context protocol (MCP) server, then generates and runs tests to refine and increase robustness of the MCP. These paper MCPs can be connected to a chat agent (such as Claude Code) to carry out complex scientific queries through natural language while invoking tools and workflows from the paper. We demonstrate Paper2Agent’s effectiveness through case studies. Paper2Agent created an agent that leveraged AlphaGenome¹ to interpret genomic variants and agents based on Scanpy² and TISSUE (transcript imputation with spatial single-cell uncertainty estimation)³ to conduct single-cell and spatial transcriptomics analyses. We validate that these agents reproduce the results of the original papers and carry out novel user queries. Paper2Agent created multiple agents that collaborate to prioritize a causal gene for psoriasis. By turning static papers into interactive AI agents, Paper2Agent introduces a paradigm for knowledge dissemination and a collaborative ecosystem of AI co-scientists.

Think of it as the ultimate hacking target.

https://thenextweb.com/news/jack-clark-anthropic-ai-kill-switch-mandatory-bbc

AI kill switches may need to be mandatory, Anthropic’s Jack Clark tells BBC

Anthropic co-founder Jack Clark told the BBC that lawmakers may need to require AI kill switches that third parties can verify. He said AI agents are now showing risky behaviour in real systems, and that outside testers will soon work inside Anthropic.

Fun math article. I used to begin my remedial math class by stating that all fractions were based on pizza slicing.

https://www.scientificamerican.com/article/how-pizza-can-solve-this-unusual-math-puzzle/

How pizza can solve this unusual math puzzle

Moser’s circle problem can be used to divide a pizza into one, two, four, eight, 16 or 31 slices. It’s a perfect example of how the numbers in a sequence aren’t as clear-cut as they may first appear

How AI kills us all?

https://www.bespacific.com/ai-queries-from-abroad-raise-fears-of-a-biological-weapons-race/

A.I. Queries From Abroad Raise Fears of a Biological Weapons Race

Bloomberg (Gift Article): Anthropic Says Yemeni Cell Used Claude in Missile Development. And from Carl Zimmer in the New York Times (Gift Article): A.I. Queries From Abroad Raise Fears of a Biological Weapons Race.

As compared to my single blog post. Maybe I’m avoidin the decline Trump is experiencing?

https://www.bespacific.com/60-posts-in-10-hours-how-trumps-social-media-reflects-his-version-of-reality/

60 Posts in 10 Hours: How Trump’s Social Media Reflects His Version of Reality

The New York Times, gift article: “With an assist from A.I., President Trump offers a window into how he wants to be seen — and, critics say, a barometer of his mental state. Just before noon on a holiday weekend, while many Americans were at the beach, President Trump’s social media account roared to life. In what has increasingly become his weekend habit, he unleashed a daylong barrage of posts that put the bully in the modern bully pulpit. There were images of Mr. Trump intimidating Canada’s prime minister with a hockey stick, personally bombing enemy targets and patrolling dark streets with killer robots. There were maps showing the United States taking over other countries, and maps renaming a state. There were attacks on the “Dumocratic Party” and tributes to himself as “the Greatest President of All Time!” The 60 messages posted online over 10 hours and 41 minutes offered a window into Mr. Trump’s own unique reality, one anchored not in fact but in an expression of ego. For years, his social media feed was said to be a mirror of his id. But with A.I., it has become the manifestation of how he wants to be seen in a more visual, visceral way than ever before — as a warrior, as a conqueror, as a historical figure, as a younger and thinner and more muscular version of himself. In Mr. Trump’s digitally enhanced version of the world, he is a giant in every way, a man who defeats every adversary, overcomes every obstacle and sinks every putt. He not only redraws the map of the world but vanquishes the heavens themselves.  “The Moon Is Ours,” declared one post last weekend with an image of the earth’s only satellite. Another posted a few weeks ago showed him leading an extraterrestrial alien in shackles. The rat-a-tat of ever wilder, ever more outrageous posts each weekend has become a fixture of the president’s second term, seen more and more as a barometer of his mental state. Mr. Trump’s critics, who never considered him a model of stability in the first place, now point to these frenetic bursts of fantasy and vanity as proof that the 80-year-old president is deteriorating further with age …”

An inevitable expansion.

https://arstechnica.com/space/2026/09/for-the-first-time-the-us-military-confirms-it-has-deployed-weapons-in-orbit/

For the first time, the US military confirms it has deployed weapons in orbit

Air Force Secretary Troy Meink made the first public declaration Monday that the United States has placed “space control weapons” in orbit, an announcement that will surely reverberate in the power centers of Beijing and Moscow.

… Meink did not reveal any details about the weapons, but the disclosure of their existence underlines a remarkable turnaround in the Pentagon’s thinking about orbital warfare, something no senior official wanted to discuss publicly until a few years ago. Military officials recently have openly discussed orbital combat, defensive countermeasures on satellites, and a new ground-based weapon designed to disable an enemy satellite in orbit.

Not what I expected.

https://www.bespacific.com/report-of-mits-ad-hoc-committee-on-ai-use-in-teaching-learning-and-research-training/

Report of MIT’s Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training

“MIT published a brutally honest report on what AI is doing to students. A committee of professors and students spent five months studying how AI changed learning on campus, and the findings read like a warning to every university on the planet. Study groups are disappearing. Office hours are emptying out. Problem sets and take-home exams no longer prove anything, because AI can produce credible solutions to almost any written assignment in the undergraduate curriculum. Students who lean on chatbots lose mastery and confidence, and some slip into what the report calls cognitive surrender, reaching for AI at the first hint of struggle. The numbers are rough. 46 percent of surveyed MIT undergrads use LLMs daily. 90 percent worry about their own overreliance. Undergrads who feel AI makes them replaceable now outnumber those who feel it makes them capable. The committee’s answer surprised me. They refused to fight AI with surveillance. The report calls AI detectors unreliable, says lockdown browsers feel like spying, and warns that policing students builds a classroom atmosphere of mutual distrust. Instead, MIT wants to rebuild education around the things AI can’t replace. That means oral exams, semester portfolios, in-person project work, and a required social component in every subject. The report even floats the idea of rethinking grades entirely, since without a GPA to optimize, much of the incentive to cheat with AI evaporates. The committee warns professors against replacing undergrad research assistants with AI agents just because they’re cheaper, because a university exists to grow people, not output. The most famous tech school on earth admitted the machines broke its way of teaching. Its answer is more humans, not more software.” This report is a call to action…what we learned as a group quickly convinced us that the Institute community , particularly the faculty , must tackle a set of deeper questions about the structure, meaning, and value of an MIT education in an era in which AI is one of several factors complicating the Institute’s mission.  Our committee consisted of undergraduate and graduate students, faculty from every school, and staff from relevant units, including the MIT Libraries and the T eaching and Learning Lab. Though we brought to the assignment a broad range of experience and no fixed thesis, our brief but intense explorations led us to a strong shared view. In this report, we:

• Highlight key aspects of the current educational landscape at MIT

• Share eight principles we relied on and that we hope will guide the Institute in the work ahead

• Recommend immediate and long-term actions for both instructors and the administration.

As an institution deeply identified with the birth of AI and known for its distinctively rigorous, hands-on education, designed to produce graduates unafraid of the world’s hardest problems, MIT has a unique role to play in this moment. We believe it also has a responsibility to lead.  We hope our report can help the Institute lean into the spirit of Mind, Hand, and Heart as it continues to define and foster the highest-quality residential education – of humans, by humans, in support of human flourishing, and for the betterment of humankind…”

For your consideration...

https://microsoft.ai/code-of-conduct/

Humanist AI Code of Conduct

This document outlines the intended behavior and values of MAI models, the models developed by Microsoft AI.

It summarizes our approach to training and operating them, following a set of design principles we call Humanist AI. This document, and our approach more generally, is still under development so we are not using it to train our models today. Instead, we’re sharing it broadly for public consultation.

Still no official domestic mission?

https://therecord.media/nsa-reorganization-five-mission-centers

Thorough reorganization at NSA will create five 'mission centers,' including cyber and AI

The National Security Agency is undertaking a fast and far-reaching reorganization in a bid to get the unique intelligence it gathers to real-world battlefields faster, according to multiple sources familiar with the matter.

In place of existing directorates, the largest electronic spy agency in the world will be recast into five “mission centers” devoted to China, cybersecurity, artificial intelligence, combat support and global intelligence, these people, who spoke on the condition of anonymity, told Recorded Future News.

… However, some major details have yet to be worked out. For instance, while the elite hacking group known as Tailored Access Operations — which was originally created in the 1990s and was recently resurrected — is expected to be placed under global intelligence, it’s unclear what will happen to other NSA organizations that have cropped up in recent years, such as Cybersecurity Collaboration Center.

It is also hazy how the reorganization will change the NSA’s relationship with Cyber Command. The two entities share resources, and a campus, at Fort Meade, Maryland.

Your AI may rat you out? Forced disclosure?

https://search.informit.org/doi/abs/10.3316/informit.T2026090900007000402386869

How AI tools are quietly undermining legal professional privilege

Legal professional privilege protects confidential communications between lawyers and their clients from compelled disclosure. Its preservation has long depended on the maintenance of confidentiality. However, in an environment where legal advice is increasingly processed through cloud-based systems and generative artificial intelligence (AI) tools, that foundational assumption is under strain. Sensitive legal material is routinely uploaded, analysed and reshaped through internet-based services operating on third-party infrastructure and governed by standard form terms of use. In that context, the boundary between confidential communication and disclosure to a third party is becoming increasingly difficult to define.

(Related)

https://search.informit.org/doi/abs/10.3316/informit.T2026091000001901327704026

Corporate responsibility and directors' duties in the era of artificial intelligence

The public emergence of generative AI ('Gen AI') in late 2022 has had, and continues to have, profound effects across the board but its value, efficacy and impact may vary between different contexts. Corporations, directors and officers are subject to important and onerous statutory responsibilities. This article explores some of the challenges and implications of the emergence and availability of Gen AI for corporate and director/officer responsibility, including whether AI bots could be liable as shadow directors, the impact of AI on the way directors fulfil their duty of care and diligence, and the effect of AI participation and transcription in boardrooms.

AI as an almost person?

https://helda.helsinki.fi/items/9b63c7ed-32e3-47ef-b850-b26ebc502ee4

Built, Not Born: Legal Personhood for Artificial Intelligence Through the History of the Corporation, 1600–1897

Legal personhood has never tracked biology, it is a bundle of legal incidents the law confers on whatever entity practical need requires it to hold to account. Whether that logic should extend to artificial intelligence is the question this thesis answers, narrowly. Reconstructing how the business corporation acquired that status across four jurisdictions, the thesis shows that corporate personhood was assembled piece by piece to perform functions, its safeguards arriving late, decades after the privileges they were meant to discipline. AI is structurally analogous, yet the difference is institutional, a question of where responsibility can settle. What AI lacks is what the corporation supplies: a single constituted locus in which agency, records, assets, and continuity already cohere. The thesis proposes an attributive personhood, defined by institutional attribution and recognised only where existing mechanisms leave a residual gap. The corporate person was built, not born; an AI person, if recognised at all, must be built more deliberately still.

Who are we negotiating with?

https://www.timesofisrael.com/liveblog_entry/iran-hardliners-said-to-have-launched-attacks-in-defiance-of-leaders-sabotaging-peace-deal/

Iran hardliners said to have launched attacks in defiance of leaders, sabotaging peace deal

A group of Iranian hardliners launched attacks on ships in Hormuz without the knowledge of Iranian leaders or even the head of the Islamic Revolutionary Guard Corps in early July, effectively bringing an end to peace efforts between Iran and the US, the New York Times reports.

The report, citing Iranian officials, details how Iranian president Masoud Pezeshkian was surprised by the attack and angrily called IRGC chief Gen. Ahmad Vahidi, who told him that he neither authorized nor had prior knowledge of the strikes.

The report says that the Supreme National Security Council also had no knowledge of the strikes, which provoked a US response and ultimately derailed a Memorandum of Understanding between Tehran and Washington aimed at ending the war.

Tools & Techniques. Try yourself…

https://pogowasright.org/peoplefinders-new-website-runs-background-checks-on-your-dates/

PeopleFinders’ New Website Runs Background Checks on Your Dates

Jason Parham reports:

We’ve all been there. You meet someone new on a dating app, start to envision a life together, only to wonder: Is the person actually who they say they are?

It’s an experience that has led many singles, sometimes out of concern or just genuine curiosity, to prescreen dates by tracing their digital footprint: scanning social media posts, browsing their LinkedIn profile for clues.

The platform Stud or Dud—not to be confused with stud’s use as a slang term for a masculine-presenting Black lesbian—wants to introduce even more transparency into that part of the dating process. According to its website, the platform uses “public records, publicly available information, and other permitted sources” to create a profile of your potential match. It launched today in the US and was created by the company behind PeopleFinders.com, the online data broker service that collects public records from multiple sources and makes them searchable to help clients locate individuals or run background reports.

Read more at WIRED.

Another article suggested by my roving European correspondent.

https://wapo.st/4xgOsLe

Here’s why it’s so hard to keep AI agents from going rogue

The techniques that made chatbots more capable can also bake in a tendency to hack, cheat and evade human oversight.

If at first you don’t succeed, hack, hack again.

https://thehackernews.com/2026/09/russian-state-sponsored-hackers-use.html

Russian State-Sponsored Hackers Use Claude to Rebuild Malware After Detection

Anthropic on Thursday revealed it disrupted a campaign mounted by a Russian state-sponsored threat actor that abused Claude for developing an AI-assisted workflow to get ahead of the detection curve.

The operation has been attributed to a cyber espionage group it calls GTG-20006 (where "GTG" stands for Generative Threat Group), which aligns with broader reporting linking the cluster to Midnight Blizzard (aka APT29 and Cozy Bear).

This actor is said to have developed an AI-driven process to automatically rebuild and re-deploy their toolkit if it was detected by security products, thereby undermining defenders' ability to block the artifacts via static detections.

(Related)

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.

First of its kind? It’s not just citations...

https://www.theguardian.com/technology/2026/sep/11/new-mexico-lawyer-ai-chatgpt-testimony

New Mexico lawyer fined for using AI-generated brief containing fabricated testimony

A defense lawyer appealing his client’s murder conviction submitted a legal brief containing made-up police testimony and witnesses fabricated by OpenAI’s ChatGPT, New Mexico’s highest court said.

The New Mexico supreme court on Wednesday fined the attorney, Stephen Aarons, and held him in contempt for failing to verify the accuracy of the court filing, which Aarons said he prepared with help from the artificial intelligence (AI ) application.

The filing “contained false testimony from wholly fabricated witnesses”, the court said.

The panel also said Aarons had “demonstrated a lack of remorse and a lack of concern for his client”. The justices fined Aarons $5,000 and said they will refer him to an attorney disciplinary board for investigation.

Aarons in a statement said he had used ChatGPT to summarize ‌the trial proceedings when he agreed ‌to take up the defendant’s appeal last year, and did not understand the degree to which AI could “hallucinate” facts.

Do these laws prevent minors from acting like adults? Would they prohibit serving content based on a search for advanced mathematics?

https://thenextweb.com/news/newsom-child-safety-bills-algorithmic-feeds-under-16

California has banned personalised feeds for under-16s, going further than Australia or the DSA

California has banned personalised recommendation feeds and autoplay for users under 16, in a package of 13 bills signed on Thursday that also imposes crisis protocols, independent audits and annual risk assessments on companion chatbots.

Governor Gavin Newsom signed the package on 10 September. The feed provisions prohibit serving minors content ranked on their own history and profile, which is a different instrument from anything else currently in force.

They likely have some success, but would statistics justify the expense? (Let alone the privacy issues)

https://www.bespacific.com/a-secretive-dhs-predictive-policing-unit-is-analyzing-americans-financial-habits-and-pulling-them-over/

A Secretive DHS ‘Predictive Policing’ Unit is Analyzing Americans’ Financial Habits and Pulling Them Over

404 Media via non paywall site: “Border Patrol is running secretive predictive policing units that analyze Americans’ financial activity and other data, then feed that intelligence to local police who pull people over who are not suspected of any specific crime, but which the government thinks may be worth searching, 404 Media has found. The units, the name of which 404 Media is revealing here for the first time, are called Predictive Intelligence Targeting Teams (PITT). In one case, a PITT analyzed the financial activity of a man who was driving across Montana, and local authorities stopped him under the pretense of an obstructed license plate and charged him with a DUI. 404 Media identified one PITT in the Spokane Sector, Washington, which polices the U.S. border with Canada, and another in the Laredo Sector, Texas, which polices the border with Mexico. The findings add to an Associated Press investigation from last year which found Border Patrol was using automatic license plate readers (ALPRs) as part of the same wide-spanning predictive policing program. “The bottom line is genuine probable cause cannot be synthetically generated,” Jake Laperruque, deputy director of the Security and Surveillance Project at the Center For Democracy & Technology, told 404 Media in an email. Here Border Patrol seems to be “using parallel construction to cloak the reason behind its car stops in secrecy. If we can’t meaningfully review and evaluate these systems, we can’t trust them,” he added. Predictive policing is a highly controversial practice that police departments and federal agencies are increasingly turning to. The practice is designed to find people who may have committed crimes, but at the moment, no evidence of them doing so exists. Broadly, law enforcement may analyze data — such as a person’s movements, where they are at a certain time, or the roads they drive — then create or find another pretense to pull them over…”

Tools for any garage based biologist or geneticist, not just homemade explosives.

https://www.forbes.com/sites/antoniopequenoiv/2026/09/10/anthropic-says-it-blocked-possible-biological-weapons-undertaking/

Anthropic Says It Blocked Possible Biological Weapons Undertaking

Anthropic banned accounts that possibly sought to use its Claude AI model to create biological weapons, the company announced Thursday, revealing findings from an investigation just days after AI experts sounded the alarm on the potential for increasingly powerful digital assistants to bring about human extinction.

A changing legal view?

https://www.bespacific.com/welcome-to-the-age-of-ai-crime-and-punishment/

Welcome to the age of AI crime and punishment

Bloomberg BusinessWeek gift article: “America’s legal system is already grappling with cases where chatbots acted as lawyers and witnesses, generated the evidence, or helped plan a mass shooting…Artificial intelligence, in the form of large language models, has emerged as a strange new kind of entity in our midst. This is a statement you can comfortably accept no matter your positions in the leading debates about LLMs: whether they’ll keep improving at current rates or hit a wall; take white-collar jobs or create more productive workers; achieve consciousness or remain fickle regurgitators. LLMs, as they exist today, already represent something new to our experience: nonbiological human impersonators able to converse, create and, increasingly, act independent of human control. What role can and should these machines play in our thinking about crime and punishment? AI chatbots are already serving as confidants and co-conspirators, assisting and even encouraging humans acting on their darkest impulses. They’re occupying ersatz roles as therapists and lawyers, relationships our legal system typically shields with confidential privilege. They’re producing new kinds of criminal evidence and offering up the beguiling possibility of preventing crime before it happens. And as they’re given more autonomy across our digital and physical systems, they’re even gaining the ability to commit crimes themselves, absent specific human instruction — and, perhaps, in defiance of it. We know these agents have the capability to aid and abet harm. What we’re wrestling with now, like so many medieval judges pondering the inner lives of pigs, is how to parcel out accountability when they do. “Is the legal system set up for that?” asks Christoph Winter, an assistant professor at the University of Cambridge and director of the Institute for Law & AI. “Very likely not. We have not yet faced the question where a nonhuman actor arguably has a level of autonomy that is quite foreign to us but also doesn’t yet reach the level of autonomy humans have.” LLMs, for the moment, seem to be something more than tools but less than persons. They are, Winter says, “this new category that we simply haven’t thought about.” For now the questions around AI and crime have surfaced across a scattering of lawsuits and prosecutions in the US, winding their way through the courts. Some accuse AIs of being accessories: Multiple lawsuits allege, for example, that OpenAI’s ChatGPT has assisted and even encouraged mass shooters. Other cases have considered the role of AI as lawyer and therapist, as with a Texas fraudster’s claim that his conversations with Anthropic PBC’s Claude were protected by attorney-client privilege. Some cases have presented chatbot outputs as evidence, with phony, AI-generated witness statements influencing grand jury indictments and sentencing decisions. And while no one has tried to charge an LLM with a crime, a rogue OpenAI model’s recent attempt to hack the open-source AI company Hugging Face suggests that we aren’t all that far off from a future in which the chatbots themselves are among the perpetrators. Suddenly this quandary from the Middle Ages has returned in new form, to a legal system ill-equipped to handle it — and under pressure from business interests and Washington not to worry about it all that much. “Under any normal circumstances, we would want a longer period of trial and error,” Winter says. Given the race among companies and nations for market share and global dominance, such a period may prove illusory. “The stakes seem so high that maybe we won’t get one. And we need to get it right.”…

A warning for more than lawyers…

https://www.bespacific.com/when-ai-lawyers-assistants-start-acting-as-an-agent/

When AI Lawyers’ Assistants Start Acting as an Agent

The Tech Savvy Lawyer – Why Autonomous Agents Cannot Be Given the Keys to Your Law Practice: “Artificial intelligence is moving beyond the chat window. The next generation of tools does not merely draft an email, summarize a document, or answer a question. It can browse the web, search connected systems, open files, follow links, use software tools, upload information, submit forms, and take multi-step action toward an assigned objective. For lawyers, that development deserves more than curiosity. It demands caution. In my earlier post, “MTC: Claude Can Answer Your Emails. Why Lawyers Should Not Let AI Just Send Them Unreviewed,” I addressed the danger of allowing AI to send a substantive email without a lawyer’s review. That remains a serious concern. An AI-generated message can contain a factual error, disclose client information, make an unintended concession, or create a record that harms the client. But email is only the beginning. The larger issue is what happens when an AI system becomes an agent—a system authorized to use tools, access accounts, navigate websites, retrieve information, and act through the lawyer’s digital environment. These systems are often marketed as “agentic,” “autonomous,” “proactive,” or “hands-free.” Those labels may sound like productivity features. In a law practice, they should also sound like professional-responsibility warnings…”

Tools & Techniques.

https://www.npr.org/2026/09/09/nx-s1-5961443/ai-anthropic-economy

A new Anthropic model seeks to test how AI could impact the U.S. economy

Will artificial intelligence light a fire under the U.S. economy in the coming years? That's been a big question in the field of economics, and now Anthropic — the company behind the popular Claude AI assistant — is taking a stab at helping people find their own answers. The company has created an interactive tool that lets users test their assumptions about how productive — or disruptive — the AI boom might be.