Anthropic says it blocked five potential bioweapon plots
The September 11, 2026 edition of MIT Technology Review's daily newsletter, The Download, leads its must-read roundup with Anthropic's disclosure that it has blocked potential plots to build biological weapons, identifying five such cases, plus six cases of people trying to use AI to build software for conventional weapons. The newsletter, citing outside reporting on the same disclosure, adds that governments are also using Anthropic's Claude for surveillance and that Russia-linked hackers used it to automate attacks on Ukraine; it describes all of these as threats revealed in a new Anthropic report. A separate linked piece has Bill Gates arguing that AI needs new guardrails.
The issue also carries two of the newsletter's own features, both tied to MIT Technology Review's 2026 list of 35 Innovators Under 35. One, bylined by Bridget Reed Morawski, profiles Laureen Meroueh, founder of Hertha Metals, who may have found a way to clean up steelmaking without raising its price. Her new furnace turns iron ore into refined liquid steel in a single step and swaps coal for natural gas, which she says cuts emissions by at least half and costs by 25% compared with conventional steelmaking; the newsletter notes that the industry's underlying purification process has barely changed since it was invented and commercialized in the 1850s. The other feature, by Jessica Hamzelou, introduces the list's biotech category. Biotech is one of four categories in the 2026 list, alongside AI, computing and robotics, and climate and energy, and nine of this year's honorees fall into it. Their work, as described in the newsletter, spans a reprogramming therapy that reverses vision loss, tiny brain electrodes inspired by Japanese art, a personalized gene-editing treatment for a baby with a rare genetic disorder, and efforts to design new viruses with generative AI that could, hopefully, produce new drugs or help clean up pollution.
The rest of the must-read list ranges widely. California has banned addictive social media features, including infinite scroll and autoplay, for users under 16; it is the first law of its kind in the US, and it also sets new rules for AI companion chatbots, which the newsletter links to wider AI fears pushing Congress toward tougher regulation. Two AI researchers have left Anthropic and Google over safety risks, a day after what the newsletter calls Jacob Coxon's viral departure from Anthropic, and Elon Musk dismissed their concerns as a "setup" and a "psyop." Sam Altman is pitching OpenAI's cyber defenses to power companies after reports of AI attacks on critical systems, and separately told OpenAI staff that the company is open to slowing down its AI work. After years of resisting AI, music labels are starting to embrace it: Universal is partnering with ElevenLabs on an AI remix platform. Chinese drugmakers are developing hundreds of GLP-1 treatments as they challenge US dominance in weight-loss drugs. Electric air taxis have begun official test flights in Texas under a new White House pilot program. Chinese drones are helping deliver food and airlift bodies in flood-hit parts of Nepal. NASA and IBM have built an AI model to map the moon that could help locate ice and identify safer landing sites. The list closes on a lighter note: one man's Restroom Archive is a museum-style repository of 3D scans of America's public restrooms.
The newsletter's quote of the day comes from Kalie Roberts, a travel content creator, who says in an Instagram reel that Meta AI pieced together her children's identities and her family's location from years of her Facebook posts, including a video of her singing in the car with her kids; the newsletter does not report any response from Meta to her claim. Its closing feature, by Caiwei Chen, describes Chinese tech workers whose employers are pushing them to document their own workflows, using tools such as OpenClaw, so that work can be automated. The trend followed a spoof GitHub project called Colleague Skill that surfaced in April and claimed to distill a worker's skills and personality into an AI agent; it left many workers fearing they are being flattened into code, and some are now building tools meant to sabotage the automation process. A short closing section of lighter items includes a car designed to fool Flock's surveillance cameras, a new Webb Space Telescope image of a galactic merger, a 66-million-year-old fossilized bird feather found preserved in a fossilized dinosaur dropping, and a preservationist who traveled 1,700 miles and made 52 calls to keep a rare phone box in service.
Key facts
- Anthropic says it identified and blocked five potential plots to build biological weapons and six cases of AI-built software for conventional weapons; the same disclosure, per the newsletter, found governments using Claude for surveillance and Russia-linked hackers using it to automate attacks on Ukraine.
- California has banned addictive social media features such as infinite scroll and autoplay for users under 16, the first law of its kind in the US, and the law also sets new rules for AI companion chatbots.
- Two AI researchers left Anthropic and Google over safety risks a day after Jacob Coxon's high-profile departure from Anthropic; Elon Musk dismissed their concerns as a "setup" and a "psyop."
- Hertha Metals founder Laureen Meroueh says her new furnace, which turns iron ore into liquid steel in one step using natural gas instead of coal, cuts emissions by at least half and costs by 25% compared with conventional steelmaking.
- Sam Altman is pitching OpenAI's cyber defenses to power companies after reports of AI attacks on critical systems, and told OpenAI staff the company is open to slowing down its AI work.
Why it matters
This single edition shows how routine two once-exceptional things have become: a frontier AI lab publicly reporting that its own model was misused for weapons research and state-linked cyberattacks, and a piece of climate tech chasing an incremental fix to a stubborn industrial process. Anthropic's disclosure, covering bioweapon-related plots, conventional-weapons software, government surveillance and Russia-linked attacks on Ukraine, lands next to a California law curbing addictive social media design for minors and two more safety-driven departures from Anthropic and Google; together they suggest both the harms AI can enable and the political response to them are moving on the same timeline. Hertha Metals' furnace matters for a different reason, if Meroueh's numbers hold up: steelmaking is a major industrial source of carbon emissions, and its core purification process has not fundamentally changed since the 1850s, so a method that cuts both emissions and cost, rather than trading one for the other, removes the usual excuse for not switching.
Who it affects
Anthropic and its enterprise and government customers; the unnamed state-linked hackers and weapons developers the report describes; and Ukraine, as the reported target of Russia-linked attacks. Anthropic's and Google's AI safety teams, after two more researchers left over risk concerns. California's under-16 social media users and the platforms that must now redesign features like infinite scroll for them, plus any legislature watching the law's effect. Power utilities that Sam Altman is courting with OpenAI's cyber defenses. Steelmakers and heavy industry, if Hertha Metals' process scales, and this year's nine biotech honorees, along with the patients and fields their work touches, from vision loss to rare genetic disease. Chinese tech workers, per Caiwei Chen's piece, whose employers are pushing them to document their own workflows for automation.
How to use it
There is nothing here to buy or deploy yet. Hertha Metals' furnace has no stated commercial timeline in this piece, and Anthropic's report itself is not reproduced here, only summarized through outlets including the New York Times, BBC, Axios, Quartz and the Guardian; readers who want the underlying detail on the bioweapon and misuse findings need to follow those links, or Anthropic's own report, directly. The newsletter itself offers several concrete next steps: its own links to the biotech honoree profiles, to the rest of the 2026 list across all four categories, to the full Hertha Metals feature, and to the full story on the Chinese workers training their AI doubles; editorially, it also points biotech readers to The Checkup, its weekly biotech newsletter where the biotech story originated and which readers can sign up to receive.
How solid is it
Most of this edition is MIT Technology Review curating and summarizing other outlets' reporting in one line each, not reporting it firsthand, so the bioweapon, surveillance and Ukraine-attack claims carry only as much weight as those original stories do; this text does not show Anthropic's report itself. The steel figures, cutting emissions by at least half and costs by 25%, are explicitly Meroueh's own claim, introduced with "she says," with no independent verification, baseline comparison or commercial timeline given here. Musk's "setup" and "psyop" characterization is his own reading of the departing researchers' motives, not an established fact. The Meta AI claim rests on one user's account of her own Instagram reel, which this text does not show Meta confirming or disputing.
Risks and caveats
Several claims are hedged in ways worth keeping. Meroueh "may have found" a fix, not confirmed one, and the newsletter's description of designing new viruses with generative AI is framed as something that will "hopefully" produce drugs or clean up pollution, not as something already achieved. It is not clear from this text whether the two researchers who left Anthropic and Google came one from each company or both from one, nor whether Musk's dismissal targeted them, Jacob Coxon, or both. The newsletter does not name the power companies Altman met, the two departing researchers, or the specific iron-ore process Hertha Metals' furnace would replace. Treat the bioweapon and cyberattack figures as what Anthropic says it found and chose to block, not as a full account of attempted AI misuse, since a report from the company doing the blocking cannot describe what it missed.
“I didn't ask Facebook to build a profile of my family. I posted a video of me singing in the car with my kids.”
— Kalie Roberts, a travel content creator, in an Instagram reel