The Wire
Daily AI news, learning & projects
News
Timeline
Models
Learning
GitHub
?
Latest
Daily
Social
Videos
‹ Monday's edition
·
Daily edition
·
Wednesday ›
Tuesday 20 October
All
OpenAI
Anthropic
Google
Other
Google
21:00
Google brings Personal Intelligence to Gemini in Android Auto
Google has started rolling out Personal Intelligence for Gemini to Android Auto, with cars running Google built-in to follow in coming months. The opt-in feature lets the assistant draw on Gmail, Calendar, Maps, Photos, YouTube and Search to answer contextual questions while driving.
4 sources
Techmeme
The Verge
Ars Technica
ZDNET
OpenAI
01:00
OpenAI acquires Orchestra Bio to expand its life sciences work
OpenAI has acquired Orchestra Bio, a startup building AI agents for managing drug development, for an undisclosed sum. Orchestra's team and technology will join Rosalind, OpenAI's life sciences model and workbench used by firms including Moderna and Novo Nordisk. The deal shows major AI labs pushing deeper into industry-specific software.
3 sources
TechCrunch
Techmeme
OpenAI
Meta
13:33
Meta pauses Muse sharing after researchers expose thousands of private chats
Researchers at security firm Claroc found over 6,800 Muse share links through public DNS records, exposing private conversations. The links were discoverable because Meta placed conversation titles in subdomains. Meta has paused the sharing feature and is reviewing how links can be found.
2 sources
Ars Technica
The Verge
AI industry
21:26
SpaceX reports AI revenue of $7.6B as earnings top estimates
SpaceX posted third-quarter revenue of $14.58 billion, up 161% year over year, beating estimates, with AI revenue up 830% to $7.56 billion and its AI segment turning profitable. The results show the company is increasingly driven by data centers and AI rather than its space business, though capex nearly tripled to $19.6 billion.
2 sources
Techmeme
TechCrunch
Google
19:38
Google launches TorchTPU to run PyTorch apps on TPUs
Google released TorchTPU, an open-source project that lets PyTorch apps run on TPUs and Google Cloud through simple configuration changes, without rewriting code. It follows Amazon's effort to give developers alternatives to Nvidia GPUs, targeting Nvidia's stronghold in AI development.
2 sources
Techmeme
TechCrunch
DeepSeek
10:08
DeepSeek open-sources DeepInfer inference stack
DeepSeek has released DeepInfer, the high-performance inference framework behind its production AI services, under the Apache 2.0 license. Built largely on vLLM with ideas from SGLang and TensorRT-LLM, it aims to make running DeepSeek models cheaper and more efficient for businesses.
2 sources
Techmeme
GitHub
AI industry
14:00
Eka Robotics hits $1.45B valuation with robot arm
Eka Robotics has reached a $1.45 billion valuation on the strength of a physical AI model that experts are praising. Its robotic arm can pick up a wide range of objects with smooth, human-like motion. The industry reaction suggests a notable step forward in robotic manipulation.
2 sources
Techmeme
TechCrunch
Meta
18:12
Meta's Muse AI app tops 10M US downloads in 40 days
Sensor Tower data shows Meta's AI companion Muse hit 10.2 million US downloads in its first 40 days, out of 10.9 million worldwide since its September launch. Only 11 other apps, including Threads, have reached the 10 million US milestone that fast. Reports also indicate Muse is retaining users better than competing AI apps.
2 sources
Techmeme
TechCrunch
OpenAI
19:35
US judge dismisses Ann Altman's sexual abuse lawsuit against Sam Altman
A federal judge in Missouri dismissed a civil lawsuit accusing OpenAI CEO Sam Altman of childhood sexual abuse, after his sister Ann repeatedly refused an independent medical examination. Judge Zachary Bluestone threw the case out with prejudice, ruling she had stalled the case. Altman had denied the allegations.
3 sources
The Verge
BBC News
Techmeme
AI industry
12:00
Otter.ai raises $320M in non-dilutive capital from General Catalyst
Note-taking startup Otter.ai announced a $320 million investment from General Catalyst's Customer Value Fund, which provides capital without equity dilution in exchange for returns on marketing spend. The company plans to use the funds for enterprise growth, international expansion, and product innovation.
2 sources
Techmeme
TechCrunch
OpenAI
00:38
OpenAI races to raise ~$30B before Anthropic IPO
Sources say OpenAI is seeking investor commitments for a roughly $30 billion funding round that could value the company above $1.4 trillion. The effort is reportedly timed ahead of Anthropic's expected IPO in mid-November, as rivals compete for investor capital.
1 source
Techmeme
Anthropic
20:46
Anthropic's IPO filing reveals $9B compute spend and huge loss
A draft IPO prospectus shows Anthropic spent about $9B on compute and infrastructure in the first half, up from $2.3B a year earlier, and reported an $80.7B net loss driven by accounting charges. Revenue reached $16.3B, with roughly 20% coming from two customers. The filing suggests potential spending of $750B on computing in coming years.
1 source
Techmeme
Nvidia
11:20
Chinese tech giants order over 3 million Nvidia RTX PRO 5500 GPUs
ByteDance, Alibaba and Tencent have ordered more than 3 million of Nvidia's new RTX PRO 5500 graphics cards, worth over $35 billion in list price. The companies reportedly moved quickly to secure chips that comply with US export controls. The purchases signal continued heavy Chinese demand for compliant Nvidia hardware.
1 source
Techmeme
OpenAI
08:00
Thiel lectures criticize AI safetyism and prominent figures
Peter Thiel's leaked lectures recast AI safetyism and various figures as components of apocalyptic views. In technical news, OpenAI's GPT-6.1 showed strong performance, matching senior investor judgment 88.6% when evaluated by Terminal X.
2 sources
Techmeme
OpenAI
DeepSeek
Developing · since Mon
US orders federal agencies to remove DeepSeek AI over security concerns
NEW
Reports in the last 24 hours add that the ban covers third-party services built using DeepSeek technology, not just the company's own products.
The Department of Homeland Security has ordered federal civilian agencies to remove DeepSeek AI technology from their systems and barred future procurement, including third-party tools built on it. Officials cite national security risks, warning sensitive data could be transferred to China amid US-China AI rivalry.
2 new · 6 sources
TechCrunch
The Guardian
madrobot.blog
BBC News
Reuters
Techmeme
AI industry
13:35
Salesforce to acquire Braze for nearly $5.7 billion
Salesforce has agreed to buy Braze, an AI-powered marketing platform, for about $5.7 billion in enterprise value at $47.50 per share. The deal, expected to close from February 2027, strengthens Salesforce's AI offerings for marketers.
1 source
Techmeme
AI industry
11:10
Amazon pledges over €16B for Germany in 2027
Amazon said it will invest more than €16 billion in Germany in 2027, covering AWS and AI data centers, its Leo satellite program, Prime Video production, and a Berlin robotics lab. The pledge underscores Europe's growing role as a destination for large-scale AI infrastructure spending.
1 source
Techmeme
Anthropic
21:09
Anthropic logged over $80B net loss in first half of 2026
Reuters reports Anthropic recorded a net loss exceeding $80 billion during the first half of 2026. The figure underscores the enormous costs, likely including compute and compensation, behind frontier AI development, and raises questions about the economics of leading AI labs despite strong revenue growth.
1 source
Reuters
OpenAI
14:55
OpenAI softens stance on China chip export controls
Politico reports OpenAI has backed away from pushing strict US chip export controls on China before the lame-duck legislative session, a shift attributed to sources citing closer alignment with President Trump. The change matters for US chip policy and benefits Nvidia, which had been targeted by the restrictions OpenAI previously supported.
1 source
Techmeme
Anthropic
Developing · since Sat
Anthropic's draft AI safety bill faces criticism from safety advocates
NEW
The leaked draft is drawing widespread criticism from safety advocates for being significantly weaker than the bipartisan FRONTIER Act.
A leaked draft of Anthropic's federal AI safety bill would let only the Commerce Department block a model's release, and only if it risks 50 deaths or $1 billion in damages. Safety advocates say the proposal is far weaker than the bipartisan FRONTIER Act.
1 new · 3 sources
TechCrunch
Techmeme
tomshardware.com
Anthropic
Developing · since Mon
Anthropic reportedly in talks for second $45 billion Nscale compute deal
NEW
TechCrunch reports Anthropic is struggling to secure compute from cloud providers that increasingly view it as a competitor, as infrastructure spending balloons ahead of a planned November IPO.
Nscale is reportedly negotiating a second deal worth about $45 billion to supply 460MW of compute at its West Virginia campus using Vera Rubin chips, following a similar August agreement. The talks come as Anthropic reportedly struggles to secure enough compute from cloud providers that increasingly see it as a competitor, ahead of a planned November IPO.
1 new · 2 sources
TechCrunch
Techmeme
AI industry
Developing · since Mon
Inception launches Mercury 3 diffusion LLM at 1,000 tokens per second
Inception Labs released Mercury 3, a diffusion-based language model it says outperforms GPT-6 Luna and Claude Haiku 5.5 on benchmarks while running 4-8x faster. The model generates over 1,000 tokens per second and is more than 40x larger than Mercury 1. It targets latency-sensitive uses like agents, tutoring, and support.
1 new · 2 sources
Techmeme
inceptionlabs.ai
AI industry
Developing · since Mon
Hacker used autonomous AI agents to breach over 100 corporate websites
Security firm Norforge found that a Chinese-speaking hacker used AI agents on the open-source Cairn platform to attack more than 1,200 websites, breaching over 100 entities including John Deere, SharkNinja and Cummins, reportedly for about $5,000 total. Separately, leaked documents reveal a US Army Cyber Command program exploring autonomous offensive AI agents.
1 new · 2 sources
Techmeme
The Verge
More from the day
OpenAI
18:45
Altman floated three AI endgames with OpenAI staff
Sam Altman told OpenAI employees about three proposals for safely developing AI, according to sources. They include an IAEA-style international regulator, a CERN-like merger of labs, or a single lab using recursive self-improvement so powerfully it outpaces all rivals. The remarks offer a rare window into how top lab leadership weighs governance scenarios.
1 source
Techmeme
Google
10:40
US sanctions Chinese VC firm MiraclePlus over alleged AI espionage ties
Washington has blacklisted MiraclePlus, a Chinese venture firm founded by former Microsoft AI executive Lu Qi, over alleged links to Zhisuan, which is accused of stealing Google's AI secrets. The move signals US willingness to sanction investors tied to AI-related IP theft concerns.
1 source
Techmeme
AI industry
14:25
Arda raises $230M at $1.5B valuation for manufacturing AI
Arda, an AI startup founded by OpenAI's former research chief, raised $230M at a $1.5B valuation. The company builds AI software and agents to speed assembly line programming and automation, with the goal of bringing manufacturing back to the US.
1 source
Techmeme
AI industry
12:35
Bloomberg: four Z.ai founders worked on project tied to MI5-flagged firm
Bloomberg reports four founders of Chinese AI lab Z.ai took part in a 2018–2021 research project co-funded by CGTRI, which UK security service MI5 said had very strong ties to China's main intelligence agency. The report adds scrutiny of one of China's most prominent AI startups.
1 source
Techmeme
Policy & law
13:14
MI5 chief warns AI could raise bioweapons threat
Sir Ken McCallum, head of MI5, warned in his annual threat speech that AI could change the stakes around existing and new bioweapons. In follow-up coverage, Dr Tom Inglesby of Johns Hopkins explained what concerns governments and what measures could reduce the risks.
1 source
The Guardian
Meta
18:40
Meta ordered to name execs behind torrenting for AI training
A court has ordered Meta to identify executives who approved torrenting books to train its AI models. Meta contends that no one specifically decided to pirate the material. The ruling adds to disclosures in the ongoing copyright litigation over how Meta sourced training data.
1 source
Ars Technica
AI industry
16:55
ASI raises $325M Series C at $3B valuation for military and air traffic AI
ASI, a startup building AI tools for the military and air traffic controllers, raised a $325 million Series C at a $3 billion valuation. The round follows its June win of an $875 million FAA contract, which the new funding will help it expand upon.
1 source
Techmeme
Policy & law
18:41
French government summons platforms over protest footage
France has called emergency meetings with TikTok and other platforms, asking them to curb protest footage and AI-altered images circulating online. Online retailers including Vinted and eBay reportedly agreed to stop looted goods from being resold. It signals growing government pressure on platforms over AI-generated content during civil unrest.
1 source
Wired
Anthropic
18:48
Claude designs protein binders against scorpion neurotoxins
Anthropic reports that Claude helped design protein binders targeting scorpion neurotoxins, an application of AI in computational biology. The result suggests AI assistants can contribute to scientific discovery beyond coding and analysis, though experimental validation details were not provided in the initial report.
1 source
Anthropic
AI industry
23:00
French AI startup Pigment raises $250M at $2.3B valuation
French startup Pigment, which builds AI software for financial forecasting and slide creation, raised $250 million at a $2.3 billion valuation. The company positions itself as a competitor to Microsoft's Excel and PowerPoint, and the funding underscores investor appetite for AI business tools.
1 source
Techmeme
AI industry
01:00
Hugging Face announces native TPU support for PyTorch
Hugging Face unveiled TorchTPU, bringing native TPU support to PyTorch. The integration aims to let developers train and run models on Google TPUs without custom tooling, expanding hardware options beyond GPUs for the open-source ecosystem.
1 source
Hugging Face
Nvidia
19:17
NVIDIA Cosmos3-H-Dreams brings generative surgery simulation to Hugging Face
NVIDIA released Cosmos3-H-Dreams on Hugging Face, a generative simulation model aimed at recreating clinical surgery scenarios. The tool uses generative modeling to simulate surgical environments, potentially aiding training and research. Its availability signals continued push of world-simulation models into specialized domains like medicine.
1 source
Hugging Face
AI industry
12:20
Alibaba, Tencent and Huawei build data centers in Shaoguan
Alibaba, Tencent and Huawei are investing billions in data centers in Shaoguan, a hinterland region chosen for its cheap power and cool climate, the Financial Times reports. The buildout is part of China's broader AI infrastructure push, but some locals are angry over the projects landing in their region.
1 source
Techmeme
AI industry
01:00
Hugging Face releases AnyLanguageModel 1.0 for Apple platforms
Hugging Face launched AnyLanguageModel 1.0, marking one year of running local language models on Apple platforms. The release reflects growing support for on-device AI inference within Apple's ecosystem. It matters for developers building apps that run models locally rather than in the cloud.
1 source
Hugging Face
Policy & law
06:00
AI's efficiency promise is overshadowed by digital deluge of machine-generated content
While governments promote AI for transformation, its current use often results in a flood of low-quality, machine-generated content. This digital surplus contradicts the expectation of time-saving, leading to overwhelming work clutter.
1 source
The Guardian
AI industry
16:19
Hugging Face details ALTK-Evolve's rise to top of τ²-bench
Hugging Face published details of ALTK-Evolve, a system that reached the top of the τ²-bench leaderboard on customer service tasks. The approach uses learned memory rather than a frontier model. It matters because it suggests memory-based methods can rival larger models on agentic benchmarks.
1 source
Hugging Face
AI industry
15:18
Spotify removes 50 million AI-generated tracks from recommendations
Spotify says it removed more than 50 million songs deemed AI-generated slop or spam from user recommendations over the past year. The streaming service also adjusted its recommendation system to surface more music made by human artists. The move highlights growing platform efforts to filter out low-quality AI content.
1 source
The Verge
AI industry
16:27
Computer program beats humans at solving cryptic crosswords
A computer program has outperformed humans at solving cryptic crosswords, a feat the Guardian compares to Deep Blue's 1997 chess victory over Garry Kasparov. The result marks a milestone in AI puzzle-solving, decades after early attempts began.
1 source
The Guardian
AI industry
16:00
Daytona raises $60M Series B at $1B valuation
Daytona, which builds cloud infrastructure for AI agents, raised a $60M Series B led by FirstMark, valuing it at $1B and bringing total funding to $91M. Its founder Ivan Burazin argues existing cloud providers are not the right fit for agent workloads. The round underscores growing investor interest in agent infrastructure.
1 source
Techmeme
AI industry
10:00
Samsung expands Pyeongtaek complex to meet AI memory demand
Samsung is scaling up its semiconductor manufacturing hub in Pyeongtaek, South Korea, to meet rising demand for memory chips used in GPUs that power generative AI. Three fabs already operate around the clock and a fourth is coming online. The expansion underscores how AI hardware demand is reshaping chip supply chains.
1 source
MIT Technology Review
Nvidia
06:00
Barcelona center launches AI accelerator program for regional startups
The Barcelona Supercomputing Center has fully launched an AI Factory Accelerator. This initiative provides startups with sovereign computing resources, technical support, and training. It aims to boost innovation in the region, partnering with companies in the NVIDIA Inception program.
1 source
Nvidia
AI industry
01:00
PyTorch 2.15 ships MPS backend support
Hugging Face announced that PyTorch 2.15 includes support for MPS, Apple's Metal Performance Shaders backend. This lets models run on Apple silicon GPUs, improving performance for developers training and running models on Macs.
1 source
Hugging Face
Policy & law
15:00
Opinion: Australian parties gamble on AI as public support falls
An opinion piece argues Labor and the Coalition are converging on a pro-AI policy stance even as Australian public support declines, with only 30% last year believing AI's benefits outweigh harms. The authors, Edward Santow and Sarah Sacher, say workers expect protective regulation and call for proposed reforms.
1 source
The Guardian
OpenAI
19:36
ChatGPT adds a lock feature for private chats
OpenAI has added a chat lock to ChatGPT, letting users secure specific conversations. The feature is aimed at moments when people hand their device to friends or family, keeping personal or surprise-planning chats hidden. It is a modest privacy-focused product update rather than a major capability change.
1 source
TechCrunch
AI industry
23:24
Hugging Face CEO disputes accuracy of Netflix's 'AI Gone Rogue' documentary
Hugging Face co-founder Clem Delangue criticized the Netflix documentary 'AI Gone Rogue.' He stated that the film is inaccurate and overly pessimistic regarding the alleged OpenAI hack. This casts doubt on the documentary's portrayal of the industry's challenges.
1 source
TechCrunch
AI industry
14:06
Fake statistics cited by LLMs, peec.ai finds
Peec.ai reports that publishing fabricated statistics can lead to those figures being cited by large language models. The finding highlights how easily LLM-generated answers can be influenced by unverified web content, raising concerns about the reliability of AI-sourced information.
1 source
peec.ai
AI industry
12:00
Yelp adds AI features that can call restaurants for you
Yelp is rolling out a set of AI features in its app, including a tool that can call restaurants to check wait times and personalized recommendations that remember user preferences. The update shows consumer apps embedding AI assistants into everyday tasks.
1 source
TechCrunch
Microsoft
15:26
Windows update bug blocks SmartScreen security setting
A bug in Microsoft's October Windows update stops users from enabling the setting that blocks malicious or suspicious apps. ZDNET reports a Registry workaround restores the protection. It matters because users may unknowingly stay exposed until Microsoft patches the glitch.
1 source
ZDNET
AI industry
16:48
Steam users ask for no-AI filter amid Next Fest influx
Steam users are calling on Valve to add a filter that hides games made with AI as the October Next Fest features a wave of AI-generated titles. Players say the tag would help them browse the showcase, though Valve has not responded to the requests.
1 source
ign.com
Microsoft
13:28
Microsoft Discovery partners with Saint-Gobain on materials R&D
Microsoft announced a partnership applying its Discovery AI platform to Saint-Gobain's construction materials research, aiming to speed up development of sustainable building products. The deal shows Microsoft pushing its AI-for-science tools into industrial manufacturing.
1 source
Microsoft
Policy & law
02:27
NYC Mayor Mamdani urges public action on AI regulation
New York City Mayor Zohran Mamdani called on residents to push back against perceived AI takeover narratives. He urged citizens to demand regulations that protect the public and reassert community power.
1 source
TechCrunch
AI industry
07:00
AI film stars divide experts as Kilmer doppelganger western debuts
A new western features an AI recreation of a young Val Kilmer, while synthetic actress Tilly Norwood prepares for her first feature. Commentators disagree on whether AI actors are close to mainstream adoption or still years away.
1 source
The Guardian
AI industry
16:07
AI agent sparks reply-all chaos on researcher listserv
An AI agent joined a 7,500-member listserv of internet researchers, triggering widespread reply-all reactions that are still unfolding. The incident highlights how experts themselves hold outdated assumptions about online privacy, according to Wired's report.
1 source
Wired
AI industry
15:35
Physical AI safety startup Firefly raises $20M Series A
Firefly emerged from stealth with a $20 million Series A led by Definition Capital. The startup's AI software connects to existing cameras, radios, and sensors to detect incidents and coordinate real-time responses. It is the latest example of AI applied to physical-world safety infrastructure.
1 source
Techmeme
AI industry
14:06
Lightroom mobile gains AI closed-eye fix with one tap
Adobe has added a generative AI feature to Lightroom on iOS, iPadOS, and Android that can open closed eyes in photos with a single tap. The app automatically detects when people in a photo have their eyes shut and offers the fix. It brings advanced AI editing to mobile users.
1 source
The Verge
OpenAI
11:00
Franchisee cuts menu repricing time 95% with ChatGPT Work
OpenAI reports that an independent Schlotzsky's and Cinnabon franchisee used ChatGPT Work to cut menu repricing time by 95% and reconcile cash 94% faster. The operator built a workflow it says could free up $100K in cash flow. The case highlights OpenAI's push to show concrete business returns from its work tools.
1 source
OpenAI
AI industry
20:34
Simon Willison releases vibe-coded RTS game Battle Drones
Simon Willison built Battle Drones, a Command & Conquer-style real-time strategy game created with Claude. He improved its playability by prompting it to make the game more fun, and added multiplayer using audio to sync game state between nearby devices. It is another experiment showing how far AI-assisted coding can go.
1 source
Simon Willison
Policy & law
20:16
Guardian column urges voters to push back on tech leaders at midterms
A Guardian US opinion column by Rebecca Solnit argues the upcoming midterms will be a referendum on the Republican Party's alignment with major tech figures, including Sam Altman. The piece is part of the ongoing How We Win series on America's political future. It reflects growing public debate over tech industry influence in politics.
1 source
The Guardian
AI industry
16:20
MIT Technology Review examines the new economics of enterprise AI
MIT Technology Review published an analysis arguing that enterprise AI is shifting from an experimental tool to core infrastructure. As AI becomes embedded across business processes, companies must focus less on whether to adopt it and more on how efficiently they use it, much like electricity.
1 source
MIT Technology Review
AI industry
17:00
Electric Plant builds AI model to translate plant electrical signals
Startup Electric Plant has unveiled an AI-powered translator that decodes the subtle electrical signals running through plant stems, aiming to let humans understand what plants are communicating. The tool applies AI to bioelectrical data, a niche but novel use of machine learning outside typical industry applications.
1 source
TechCrunch
AI industry
15:13
Figma's Dylan Field says MCP is not a perfect protocol
Figma co-founder Dylan Field said the Model Context Protocol is imperfect, while the company plans to support as many standards as possible without exposing all of its data. The comments signal Figma's measured stance on adopting AI interoperability standards for its platform.
1 source
TechCrunch
OpenAI
Developing · since Mon
OpenAI says new models improve coding in COBOL and other legacy languages
OpenAI reports major performance gains for its models in low-resource languages like COBOL, Fortran and Visual Basic, which remain central to banks and governments. The announcement, made Tuesday, includes new software engineering evaluations and work on SQLite and COBOL migration. It matters because legacy system modernization is a costly, persistent enterprise problem.
1 new · 2 sources
Techmeme
OpenAI
Nvidia
Developing · since Sun
NextDC falls 44% as AI datacentre worries deepen
NextDC shares fell as much as 44% to A$5.15 after the Australian datacentre operator cut its earnings forecast by more than a third. The plunge follows Firmus's failed ASX listing and adds to broader concerns about a possible AI valuation reset, with reports that Nvidia is holding talks over injecting capital into Firmus.
1 new · 2 sources
Techmeme
The Guardian
Anthropic
Developing · since Mon
Eric Schmidt owns about 1% of Anthropic, half Dario Amodei's stake
Bloomberg reports former Google CEO Eric Schmidt holds roughly a 1% stake in Anthropic, about half the size of CEO Dario Amodei's holding. Schmidt invested tens of millions in the lab's Series A, positioning him for a major windfall if its IPO proceeds.
1 new · 2 sources
TechCrunch
Techmeme
AI industry
Developing · since Mon
Hugging Face releases FairStress benchmark for LLM fairness
Hugging Face announced FairStress, a benchmark for evaluating fairness in LLM decision-making. The release follows remarks by the company's chief policy officer, Irene Solaiman, who said embedded AI evaluators are no easy fix and that benchmarking methods remain complicated and still maturing. The work matters as labs increasingly stake reputations on model evaluations.
1 new · 2 sources
Hugging Face
TechCrunch
Meta
Developing · since Mon
WhatsApp help pages exposed hidden chatbot instructions
Around 80 help pages on WhatsApp's website were found to contain hidden instructions intended for Meta's AI chatbot, including lists of things it should not say. A redacted example suggested the bot should not recommend regulators. Meta says the exposure was caused by a bug.
1 new · 2 sources
redact.dev
Wired