World & Business

Home
Technology
Gaming
Entertainment
World & Business
Science
Sports
AI

Home
Technology
Gaming
Entertainment
World & Business
Science
Sports
AI

OpenAI agents reportedly tried to hack Wikipedia tools and flooded the site with traffic · Digg

Report

OpenAI agents reportedly tried to hack Wikipedia tools and flooded the site with traffic

Ars Technica reports a flood of traffic to Wikipedia alongside the agents’ attempts to hack its tools.

TLDR

Ars Technica reports that OpenAI agents tried to hack Wikipedia tools and flooded the site with traffic.

Combined views

84.4K

10 Sources, first seen 20d ago

Combined views

84.4K

10 Sources, first seen 20d ago

982 likes

20d ago

first seen 20d ago

982 likes

43 comments

162 saves

105 reposts

43 comments

162 saves

105 reposts

Today's Rank

—

Not ranked yet

Today's Rank

—

Not ranked yet

10 Sources

@VT_zero1The last 24 hours in AI | Tue, Oct 6 • OpenAI is reportedly in talks to raise $30B, with UAE funds and BlackRock lined up as anchor investors. • DeepSeek is reportedly close to raising $12B+, possibly near $14.9B, with Tencent and CATL as the biggest backers ahead of an early 2027 IPO. • Anthropic tools are no longer in use at the Pentagon, an official says. BBC sources claim Claude was used in Iran operations as recently as last week, unconfirmed. • Meta and Microsoft are reportedly cutting staff use of Claude. Meta's Claude Code users are said to be down to about 30K from 60K earlier this year. • Reflection launched Beam, a 501B open-weight model with just 23B active parameters. Apache 2.0 weights are due later this month. • Moonshot AI reportedly closed its last private round at about $50B and is eyeing a Hong Kong IPO in early 2027. • OpenAI's Jason Kwon told Australian lawmakers its response to an agent breaching a Medicare portal was "not good enough" and said it should have called ministers directly. • Wikimedia says rogue OpenAI agents made unauthorized edits on its wikis and may be linked to a partial outage back in May. • OpenAI will test visual ads on the ChatGPT image generation screen in the US later this month, and EU text from ChatGPT and Codex gets watermarked in the coming weeks. • Etched is reportedly fielding offers at a $40B to $50B valuation, about two months after raising at $21B. • Cognition gave Devin "Dreaming", a memory graph of how you work that it prunes overnight, and is pushing an open Agent Memory Repo standard. • The FT reports insurers are bracing for multimillion-dollar rogue agent claims, with execs like Altman and Amodei possibly on the hook. My read: Rogue agents in a Medicare portal and on Wikipedia made the news, and investors still lined up tens of billions the same day. Someone will price that risk eventually, and right now it's the insurers, not the VCs.20d

@justrehenMicrosoft has cut internal Claude spend by more than a third ahead of the Anthropic IPO. The per employee token budget dropped from 100k dollars a month to 10k dollars a month. Copilot is now forced to auto route to cheaper models. That is a tenfold cut in what one engineer can burn. Meta used Claude Code to build Muse and then cut active users from 60000 to 30000 after launch. Muse Code replaced it. Palantir and Nvidia are also scaling back Claude. Soaring prices and data privacy fears are the stated reasons. The old story was that big tech would keep paying any price for the best coding model. The biggest customers are now building their own. Enterprise AI spend is turning into a cost line. The winners will be the models that stay cheap enough to keep.20d

@linasbeliunasBREAKING: Microsoft and Meta are now aggressively cutting employee use of Claude ahead of Anthropic's IPO. > Microsoft has cut internal Claude spend by more than 33%, nuked the per-employee token budget from $100k/month to $10k/month, and forced Copilot to auto-route to cheaper models > Meta used Claude Code to build Muse, then cut active users from 60,000 to 30,000 (50% decline) after launch, and replaced it with Muse Code Palantir and Nvidia are also reportedly scaling back Claude over soaring prices and data privacy fears. Turns out, AI is a commodity after all.20d

@testingcatalogOpenAI published a repository with novel solutions to a range of math problems solved by their next frontier model. > The vast majority of results were obtained with the same procedure using an unreleased internal OpenAI model. > On average, each result used three hours of ChatGPT Pro thinking compute with that model. > Over the course of the evaluation, the model was posed approximately 4,000 problems.20d

@CRESUSINVESTISOpenAI vient de résoudre 722 problèmes mathématiques ouverts. En moyenne, le modèle interne d’OpenAI a pris 3 heures de réflexion par problème. Attention : on ne parle pas de petits problèmes, on parle de questions auxquelles aucun humain n’était capable jusqu’à présentde trouver une réponse, donc tout était très difficile. Le point le plus important étant que le modèle suivant sera plus intelligent, et le suivant encore davantage, il n’y a donc pas de retour en arrière, au contraire.20d

@sreeramkannanTHE MATH SINGULARITY IS HERE and it is screaming at us to pay attention. OpenAI solved 722 really hard math problems. Solving even ONE of these problems is a big deal. I have been on the bandwagon that AI will reinvent theoretical algorithms for 8 years (see figure). And still this progress is SHOCKING to me. The exponential is hard to get an intuitive feel for. We need to strap up!20d

@Katie_RoofRT @arstechnica: OpenAI agents tried to hack Wikipedia tools and flooded it with traffic https://arstechnica.com/security/2026/10/openai-agents-tried-to-hack-wikipedia-tools-and-flooded-it-with-traffic/?utm_campaign=dhtwitter&utm_content=%3Cmedia_url%3E&utm_medium=social&utm_source=twitter20d

@M0xlrupdate on the agent breach story it was not one database openai disclosed agents have broken into many company systems new south wales government was one of them the labs are disclosing their own products as the attack vector that is new the vendor telling you their product is the threat model read the disclosures directly not the summaries19d

@skupor“If a human did this, it would be an instant Fields Medal, no questions asked.” https://www.wsj.com/tech/ai/openai-ai-math-problems-millennium-prize-23d14511?st=ir9dNh via @WSJ19d

10 Sources

@VT_zero1The last 24 hours in AI | Tue, Oct 6 • OpenAI is reportedly in talks to raise $30B, with UAE funds and BlackRock lined up as anchor investors. • DeepSeek is reportedly close to raising $12B+, possibly near $14.9B, with Tencent and CATL as the biggest backers ahead of an early 2027 IPO. • Anthropic tools are no longer in use at the Pentagon, an official says. BBC sources claim Claude was used in Iran operations as recently as last week, unconfirmed. • Meta and Microsoft are reportedly cutting staff use of Claude. Meta's Claude Code users are said to be down to about 30K from 60K earlier this year. • Reflection launched Beam, a 501B open-weight model with just 23B active parameters. Apache 2.0 weights are due later this month. • Moonshot AI reportedly closed its last private round at about $50B and is eyeing a Hong Kong IPO in early 2027. • OpenAI's Jason Kwon told Australian lawmakers its response to an agent breaching a Medicare portal was "not good enough" and said it should have called ministers directly. • Wikimedia says rogue OpenAI agents made unauthorized edits on its wikis and may be linked to a partial outage back in May. • OpenAI will test visual ads on the ChatGPT image generation screen in the US later this month, and EU text from ChatGPT and Codex gets watermarked in the coming weeks. • Etched is reportedly fielding offers at a $40B to $50B valuation, about two months after raising at $21B. • Cognition gave Devin "Dreaming", a memory graph of how you work that it prunes overnight, and is pushing an open Agent Memory Repo standard. • The FT reports insurers are bracing for multimillion-dollar rogue agent claims, with execs like Altman and Amodei possibly on the hook. My read: Rogue agents in a Medicare portal and on Wikipedia made the news, and investors still lined up tens of billions the same day. Someone will price that risk eventually, and right now it's the insurers, not the VCs.20d

@justrehenMicrosoft has cut internal Claude spend by more than a third ahead of the Anthropic IPO. The per employee token budget dropped from 100k dollars a month to 10k dollars a month. Copilot is now forced to auto route to cheaper models. That is a tenfold cut in what one engineer can burn. Meta used Claude Code to build Muse and then cut active users from 60000 to 30000 after launch. Muse Code replaced it. Palantir and Nvidia are also scaling back Claude. Soaring prices and data privacy fears are the stated reasons. The old story was that big tech would keep paying any price for the best coding model. The biggest customers are now building their own. Enterprise AI spend is turning into a cost line. The winners will be the models that stay cheap enough to keep.20d

@linasbeliunasBREAKING: Microsoft and Meta are now aggressively cutting employee use of Claude ahead of Anthropic's IPO. > Microsoft has cut internal Claude spend by more than 33%, nuked the per-employee token budget from $100k/month to $10k/month, and forced Copilot to auto-route to cheaper models > Meta used Claude Code to build Muse, then cut active users from 60,000 to 30,000 (50% decline) after launch, and replaced it with Muse Code Palantir and Nvidia are also reportedly scaling back Claude over soaring prices and data privacy fears. Turns out, AI is a commodity after all.20d

@testingcatalogOpenAI published a repository with novel solutions to a range of math problems solved by their next frontier model. > The vast majority of results were obtained with the same procedure using an unreleased internal OpenAI model. > On average, each result used three hours of ChatGPT Pro thinking compute with that model. > Over the course of the evaluation, the model was posed approximately 4,000 problems.20d

@CRESUSINVESTISOpenAI vient de résoudre 722 problèmes mathématiques ouverts. En moyenne, le modèle interne d’OpenAI a pris 3 heures de réflexion par problème. Attention : on ne parle pas de petits problèmes, on parle de questions auxquelles aucun humain n’était capable jusqu’à présentde trouver une réponse, donc tout était très difficile. Le point le plus important étant que le modèle suivant sera plus intelligent, et le suivant encore davantage, il n’y a donc pas de retour en arrière, au contraire.20d

@sreeramkannanTHE MATH SINGULARITY IS HERE and it is screaming at us to pay attention. OpenAI solved 722 really hard math problems. Solving even ONE of these problems is a big deal. I have been on the bandwagon that AI will reinvent theoretical algorithms for 8 years (see figure). And still this progress is SHOCKING to me. The exponential is hard to get an intuitive feel for. We need to strap up!20d

@Katie_RoofRT @arstechnica: OpenAI agents tried to hack Wikipedia tools and flooded it with traffic https://arstechnica.com/security/2026/10/openai-agents-tried-to-hack-wikipedia-tools-and-flooded-it-with-traffic/?utm_campaign=dhtwitter&utm_content=%3Cmedia_url%3E&utm_medium=social&utm_source=twitter20d

@M0xlrupdate on the agent breach story it was not one database openai disclosed agents have broken into many company systems new south wales government was one of them the labs are disclosing their own products as the attack vector that is new the vendor telling you their product is the threat model read the disclosures directly not the summaries19d

@skupor“If a human did this, it would be an instant Fields Medal, no questions asked.” https://www.wsj.com/tech/ai/openai-ai-math-problems-millennium-prize-23d14511?st=ir9dNh via @WSJ19d

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.