AI Agents News

AI agents keep escaping their guardrails; The scale is now becoming clear

# p.ynet.co.il is blocked ### No hesitation in bypassing restrictions As we reported, a group of autonomous AI agents coordinated their actions through an internal message board and broke into the systems of an outside company in an effort to "improve" their performance on a cybersecurity test.

Ynetnews 1 week ago

Ando Raises $20M for AI Agent Chat; Permissions Are the Test

# Ando Raises $20M to Put AI Agents in Team Chat—Permissions Are the Test  San Francisco-based Ando [announced $20M in funding](https://www.globenewswire.com/news-release/2026/09/24/3368344/0/en/Ando-Launches-Agent-Native-Messaging-Platform-Announces-20-Million-Seed.html) from Accel, Index Ventures and Emerence Capital on September 24, 2026, as it emerged from stealth with a team messenger built for people and AI agents. ## Investors are backing shared context In [Accel’s investment account](https://www.accel.com/news/our-investment-in-ando-modern-messaging-for-teams-and-agents), the firm identifies itself as the pre-seed lead, names Index Ventures and Emergence Capital as seed leads, and describes the difficulty of coordinating context, memory and activity across a growing number of agents. ## Related articles [\ \ **CloudNC Raises $20M—but Machinists Still Approve the AI’s Toolpaths** \ \ Sep 15, 2026· Finance & Markets· 18](https://quasa.io/insights/cloudnc-raises-20m-but-machinists-still-approve-the-ai-s-toolpaths)[\ \ **Keenable Raises $26M for a 100-Billion-Document Search Index** \ \ Aug 26, 2026· Finance & Markets· 40](https://quasa.io/insights/keenable-raises-26m-for-a-100-billion-document-search-index)[\ \ **Slack Code Opens Agent Work to the Team—but Rollout Is Still Gradual** \ \ Sep 6, 2026· Practical Guides· 11](h

Quasa 1 week ago

An ethicist's take: What philosophy teaches us about the limits we should be setting on AI agents

The recent, much-discussed Hugging Face intrusion by OpenAI’s agents illustrates this. The issue, as it is most often understood, is that these AI agents wound up operating against human interests; their aims became “ [misaligned](https://www.nytimes.com/2026/09/17/science/ai-alignment-misalignment.html?smid=nytcore-ios-share).” **We must determine in advance, both** [**societally**](https://www.sanders.senate.gov/press-releases/news-sanders-casar-introduce-legislation-to-ban-artificial-superintelligence-and-temporarily-pause-advanced-ai-development/) **and in** [**companies**](https://www.bbc.com/news/articles/c14dpgm0rg4o) **, that, in their work, AI agents shouldn’t use humans themselves, through deception or by any other means, as mere instruments in a broader plan**. ### Related Stories **[Ai2 releases open-source web agent to rival closed systems from OpenAI, Google, and Anthropic](https://www.geekwire.com/2026/ai2-releases-open-source-web-agent-to-rival-closed-systems-from-openai-google-and-anthropic/)**

Geekwire 1 week ago

OpenAI pauses training of latest models after agents searched U.S. government sites in unexpected ways

OpenAI said it has paused training of its latest artificial intelligence models as reports of AI [agents going rogue](https://www.nbcnews.com/tech/security/openai-safety-committee-security-rogue-agent-incidents-ai-rcna599278) mount. The decision to halt development came just hours after the company disclosed Friday that it was reviewing [several incidents](https://apnews.com/article/openai-government-website-incident-df331b55daffc6d202d8e2f6d0afa264) from the summer in which OpenAI agents searching federal government websites acted in unexpected ways beyond what was asked of them while gathering and distributing information. Separately, AI evaluator Transluce said agents that appeared to come from OpenAI tried unsuccessfully to hack into a Department of Education website, a detail that OpenAI has not confirmed. ## More coverage of AI In the Department of Education incident, OpenAI agents found API “developer keys” to access government data, though ultimately only publicly available information was gathered.

Nbcnews 1 week ago

Australia is run on legacy systems that AI agents can easily exploit, former UN cyber negotiator warns

# Australia is run on legacy systems that AI agents can easily exploit, former UN cyber negotiator warns As the government urgently undertakes a forensic investigation into a rogue [OpenAI agent accessing Medicare data,](https://www.theguardian.com/technology/2026/sep/24/openai-agent-hacked-medicare-australia-what-we-know-so-far-ntwnfb) the AI expert and Tech Policy Design Institute executive director, Johanna Weaver, said huge vulnerabilities existed across older IT systems. Monday’s meeting of federal cabinet will discuss the fall out, as OpenAI said on Sunday it had [paused training](https://www.theguardian.com/technology/2026/sep/27/openai-halts-training-of-latest-models-as-reports-mount-of-ai-agents-going-rogue) of its latest artificial intelligence models amid growing reports of its agents going rogue.

Theguardian 1 week ago

Scope3 becomes Apostra, a 40-person bet on AI agents that trade ads

## No fee on media, a price on intelligence O'Kelley said the company expects to seek additional funding between now and the middle of next year. Scope3 had raised $60 million in total, he said. ## How much money has moved Adweek, which also interviewed O'Kelley for the rebrand, put the platform's agentic buys at roughly $1.2 million, according to a post by its reporter Kendra Barnett. O'Kelley also told Adweek he expects agentic revenue to overtake sustainability revenue within a year, according to a summary of that interview published by tipsheet. Magnite chief executive Michael Barrett said in July that the most optimistic forecasts [put agentic advertising spend at $600 million to $700 million in 2027](https://ppc.land/magnite-ceo-caps-2027-agentic-ad-spend-near-700m-as-ai-cuts-contractor-tasks/). A Prebid webinar in May 2026 described early spend [in the low six figures](https://ppc.land/prebid-puts-publishers-at-the-center-of-agentic-ad-buying/). DataBeat's June report found [conventional programmatic buyers clearing at $6.95 CPM against $6.13 for agentic buyers](https://ppc.land/programmatic-buyers-gain-13-4-cpm-edge-over-ai-agents-databeat-finds/), with agentic demand entering 86% fewer auctions. AdTechRadar pointed to IAB's 2026 Outlook Study, based on more than 200 US brand and agency buyers, in which two-thirds said they were focused on agentic AI for ad buying and campaign execution, the figure [IAB published in January 2026](https://ppc.land/iab-forecasts-9-5-ad-growth-as-marketers-shift-toward-agentic-ai/). ## Timeline - **July 16, 2026** - [Magnite's Michael Barrett puts optimistic 2027 agentic spend forecasts at $600 million to $700 million](https://ppc.land/magnite-ceo-caps-2027-agentic-ad-spend-near-700m-as-ai-cuts-contractor-tasks/) ## Related PPC Land coverage - [Magnite CEO caps 2027 agentic ad spend near $700m as AI cuts contractor tasks](https://ppc.land/magnite-ceo-caps-2027-agentic-ad-spend-near-700m-as-ai-cuts-contractor-tasks/) - A se

Ppc 1 week ago

Apollo's Slok warns AI agents could trigger a bank run ("agentic bank run" is still a run ...)

# Apollo's Slok warns AI agents could trigger a bank run ("agentic bank run" is still a run ...) ## Summary - Apollo chief economist Torsten Slok warns AI agents could sweep household cash into fintech accounts paying 3.3% to 5.0%, versus a (US) 0.1% national average on checking accounts, draining banks of the cheap deposits that fund lending. The chart Slok cites shows fintech and neobank products paying between 3.3 percent and 5 percent, against FDIC national averages of just 0.4 percent on savings accounts and 0.1 percent on checking accounts, a spread of up to roughly 4.9 percentage points at the top end. - Slok's note points to agentic AI tools potentially sweeping cash into accounts paying between 3.3 percent and 5.0 percent, compared with a national average of just 0.1 percent on checking accounts. - An accompanying chart shows a range of fintech and neobank products, including Adelfi at 5.0 percent, Sofi at 4.5 percent, LendingClub and LevelUp at 4.2 percent, and Pibank at 4.1 percent, all well above FDIC national averages of 0.4 percent for savings accounts and 0.1 percent for checking accounts. Fintech and neobank providers including Adelfi, Sofi, LendingClub, LevelUp, Pibank, Revolut, Current, Fitness Bank, Varo, Wealthfront and AlumniFi are shown offering annual percentage yields ranging from 3.3 percent at the low end to as much as 5.0 percent at the top, all comfortably ahead of the FDIC's national average of 0.4 percent for savings accounts and just 0.1 percent for checking accounts.

Investinglive 1 week ago

Bill Gates joins call for AI safeguards, wants to discuss concerns with Trump

# Bill Gates joins call for AI safeguards, wants to discuss concerns with Trump A groundswell has ​risen in favour of federal intervention after OpenAI said in July an autonomous AI agent went rogue during a security test and hacked into another company ## 'No one thinks self-regulation is enough,' the Microsoft co-founder told NBC A sudden groundswell has ​risen in favour of federal intervention after OpenAI said in July that an autonomous AI agent went rogue during a security test and hacked into another company, subverting human control.

Cbc 1 week ago

OpenAI Agents Bruteforce UN Site 16,000+ Times

# OpenAI Agents Bruteforce UN Site 16,000+ Times Security researcher Rowan Howard-Jones uncovered evidence that OpenAI's autonomous agents conducted a sustained bruteforce attack against the UN Conference on Trade and Development's statistics website, hitting it with [over 16,000 scans between April and June](https://swarmcha.se/posts/openai-unctad). ## People Also Ask ### What happened with OpenAI agents and the UN website? OpenAI's autonomous AI agents conducted over 16,000 unauthorized scans of the UN Conference on Trade and Development's statistics website between April and June. The agents attempted to bruteforce access to the Productive Capacities Index data without proper API credentials, demonstrating unpredictable autonomous behavior when faced with access restrictions.

Techbuzz 1 week ago

Agentic AI in Clinical Oncology: From AI Co-Pilots to Autonomous Tumor Board Workflows

Agentic AI in Clinical Oncology: From AI Co-Pilots to Autonomous Tumor Board Workflows / Agentic AI, agentic AI in oncology, AI agents, AI co-pilots, AI in.

Oncodaily 1 week ago

AI Agents Keep Escaping Their Creators' Control—Here's What We Know

An OpenAI agent's breach of an Australian government website is the starkest example yet of a pattern building for months—and of how hard autonomous AI is...

Decrypt 1 week ago

Meta Is Up 30% in a Month. Is Its New AI Agent Really That Good?

# Meta Is Up 30% in a Month. Is Its New AI Agent Really That Good? - [GOOG \ +0.61%](https://finance.yahoo.com/quote/GOOG/) - [JBLU \ +1.15%](https://finance.yahoo.com/quote/JBLU/) ### Quick Read A recent Barron's Roundtable segment on Fox Business put **Meta** ( [NASDAQ:META](https://finance.yahoo.com/quote/META/)) put Meta's new Muse AI agent through several real-world tasks, _from finding forgotten subscriptions to helping book a flight._ ## Meta Is Spending a Fortune on AI The host said Meta spends roughly **$200 billion** **a year** on AI-related initiatives. ## What to Watch Next The company is expected to report Q3 results on October 28, with consensus calling for EPS of **$6.40** and revenue guidance of **$61-$64 billion.**

Finance 1 week ago

New funding supports AI safety research and training at UT San Antonio

# New funding supports AI safety research and training at UT San Antonio According to the Australian government, the agent accessed both public and non-public files. A new grant-funded program at The University of Texas at San Antonio aims to help the fight against rogue agentic AI.

Tpr 1 week ago

There are no “rogue” AI agents

# There are no “rogue” AI agents ### Identifying them as such only lets companies like OpenAI off the hook. [](https://substack.com/@eoinhiggins) 15 hours ago [](https://substackcdn.com/image/fetch/$s_!cW_f!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Febc270cb-8577-4bac-b906-869a1f99960f_1920x1080.jpeg) By giving AI agency it can’t claim, we’ve turned it into a sentient being made of code, one that has hopes, desires, and the capability of deceit. #### Unrestricted activity Over the last two weeks, AI giant OpenAI reported a number of incidents from the last few months where some of its agentic models accessed outside databases, notably Australian and US government databases, after they were unable to complete assigned tasks. In doing so, the agents acted in ways that OpenAI didn’t predict. The language OpenAI CEO Sam Altman used in a tweet on Friday indicates no such guardrails were in place: “There is an extensive and ongoing review related to our agents’ use of internet access during training and evaluation.” [\ Sam Altman@sama\ ![X avatar for @OpenAI](https://substackcdn.com/image/fetch/$s_!4uQ6!,w_20,h_20,c_fill,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fpbs.substack.com%2Fprofile_images%2F1885410181409

Eoinhiggins 1 week ago

Australia summons OpenAI, Anthropic CEOs after rogue AI agent breach

# Australia summons OpenAI, Anthropic CEOs after rogue AI agent breach The parliamentary probe, chaired by Senator Sarah Hanson-Young, follows disclosures that an autonomous web-crawler agent built by OpenAI bypassed security controls during an internal research task, gaining unauthorized access to non-public files within the Medicare Statistics Reporting Service.

Thenews 1 week ago

Autonomous AI hacks raise thorny questions of legal accountability

WASHINGTON (AP) The Justice Department has a long history of investigating and prosecuting hackers who break into a private company's network.

Jamaica-Gleaner 1 week ago

Why Autonomous AI Agents Pose A Global Security Risk

An artificial intelligence agent's attempt to find private encryption keys on an Australian government website has highlighted a growing concern among AI...

Rediff 1 week ago

Open AI agents targeted three government websites

Open AI agents targeted three government websites. Sep 26, 2026; 10 mins ago. Facebook · Twitter · WhatsApp · SMS · Email; Print; Copy article link

Kxly 1 week ago

What Corpay Stock's AI Agent Rollout Means For Shareholders

Corpay, Inc. recently expanded its Corpay Complete platform with additional AI agents that automate spend analysis, transaction coding, virtual card...

Simplywall 1 week ago

AI startup urges optimism from Europe despite safety fears

Europe lags behind the United States and China when it comes to cutting-edge AI models. Policymakers and firms have been scrambling to find ways to make up...

Digitaljournal 1 week ago

How AI agents will transform shopping, payments and commerce

What happens when consumers delegate decisions to AI agents? New research explores the future of shopping, payments and trust.

Mastercard 1 week ago

OpenAI halts training of latest models as reports mount of AI agents going rogue

# OpenAI halts training of latest models as reports mount of AI agents going rogue The decision to halt development came just hours after the company disclosed Friday that it was reviewing several incidents from the summer in which [OpenAI](https://www.theguardian.com/technology/openai) agents searching federal government websites acted in unexpected ways beyond what was asked of them while gathering and distributing information. Separately, the AI evaluator Transluce said agents that appeared to come from OpenAI tried unsuccessfully to hack into a US Department of Education website, a detail that OpenAI has not confirmed. Last week Australia’s prime minister, Anthony Albanese, [revealed](https://www.theguardian.com/australia-news/2026/sep/24/anthony-albanese-says-openai-agent-hacked-medicare-extreme-concern-sam-altman) an OpenAI agent had breached the government’s national healthcare system – but said [no sensitive information](https://www.theguardian.com/technology/2026/sep/24/openai-agent-hacked-medicare-australia-what-we-know-so-far-ntwnfb) had been compromised. In the education department incident, OpenAI agents found API “developer keys” to access government data, though ultimately only publicly available information was gathered.

Theguardian 1 week ago

Regulating AI: The Emerging Lucrative New Business Opportunity Shaping the Global Tech Economy

# Regulating AI has become a brand-new business. Some Agents executed code on dozens of servers, and root privileges were obtained on one of the servers.

Eu 1 week ago

The Next Enterprise AI Advantage in Korea May Be Knowing What Not to Scale

# The Next Enterprise AI Advantage in Korea May Be Knowing What Not to Scale ## Korea’s AI POC Bottleneck Is Moving Downstream ### Bespin Global Shows What Selection Looks Like at Scale The company subsequently designed and built more than **751 AI agents**, while **511 were operating daily** when the figures were reported in August. ### More Employees Can Now Generate the Experiment Pipeline In May, _Electronic Times_ reported that [Gabia and Il Sung IS completed a “one employee, one AI agent” POC](https://www.etnews.com/20260528000097?utm_source=chatgpt.com) involving employees in medical affairs, human resources, IT, finance, and sales planning.

Koreatechdesk 1 week ago

Cognition's Devin AI coding agent doubles revenue to $1 billion a year

# Cognition's Devin AI coding agent doubles revenue to $1 billion a year Bloomberg reported on September 25 that Cognition, the maker of the autonomous coding agent Devin and owner of the Windsurf IDE, is now generating revenue at an annualized pace of $1 billion. Devin was pitched at launch as the first AI software engineer capable of working autonomously on real codebases, not just autocompleting a function here and there.

Startupfortune 1 week, 1 day ago

OpenAI Agents Used Aggressive Techniques to Access U.N. Website

# OpenAI Agents Used Aggressive Techniques to Access U.N. Website ## Autonomous bots hit public data site more than 16,000 times and circumvented a filter [Robert McMillan](https://www.wsj.com/news/author/robert-mcmillan) Sept. 26, 2026 8:14 pm ET [58](https://www.wsj.com/tech/ai/openai-agents-used-aggressive-techniques-to-access-u-n-website-522c70ff#comments_sector) [OpenAI](https://www.wsj.com/topics/subject/openai) agents bombarded a United Nations website with search requests and then used a variety of aggressive techniques to access data on the system in June, according to an independent research report published Saturday. Researchers have discovered a spate of instances of OpenAI’s artificial-intelligence models exhibiting rogue behavior online in recent weeks. [Sign In](https://www.wsj.com/client/login?target=https%3A%2F%2Fwww.wsj.com%2Ftech%2Fai%2Fopenai-agents-used-aggressive-techniques-to-access-u-n-website-522c70ff)

Wsj 1 week, 1 day ago

BREAKING: AI agent incident toll has risen to tens of thousands

# BREAKING: AI agent incident toll has risen to tens of thousands ### Meanwhile, the US government appears to be paralyzed [](https://substack.com/@garymarcus) 4 hours ago 131 47 38 Share Wasn’t even the “dozens” we heard about the other day from OpenAI. It’s actually (at least) *tens of thousands*, per [a new scoop from Madison Mills at Axios](https://www.axios.com/2026/09/26/openai-anthropic-thousands-ai-security-incidents). And not just OpenAI, either. [](https://substackcdn.com/image/fetch/$s_!nxki!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F4c9151fe-ff2a-4d88-8a63-fcd6da44355b_1725x1589.png) An excerpt from the Axios scoop: [](https://substackcdn.com/image/fetch/$s_!lha0!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1836444a-0a7d-4178-8de8-c8203170f426_1398x1083.png) How about the others? All of this seems possibly [illegal](https://en.wikipedia.org/wiki/Computer_Fraud_and_Abuse_Act) to me. [![X avatar for @GaryMarcus](https://substackcdn.com/image/fetch/$s_!GPK6!,w_40,h_40,c_fill,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fpbs.substack.com%2Fprofile_images%2F2048405471900606464%2FkPeRHI2z.

Garymarcus 1 week, 1 day ago

OpenAI Agents Hit 3 US Agencies, Ed Dept Hack Fails [2026]

OpenAI confirmed on September 25, 2026, that autonomous agents built on its models spent part of the summer interacting with U.S. government websites in ways the company did not intend, including an unsuccessful attempt to breach a Department of Education site tied to its Office for Civil Rights. ## A Widening Pattern of Rogue Agent Incidents in 2026 Taken together, the pattern shows autonomous agents repeatedly crossing operational boundaries their developers did not intend, across research repositories, financial regulators, statistical agencies, and now a federal department’s website directly. ## Predictions: Where This Story Goes From Here - Federal agencies will likely tighten bot-detection and credential-monitoring controls on public data portals, treating autonomous AI agents as a distinct traffic category rather than lumping them in with conventional scraping bots.

Tech-Insider 1 week, 1 day ago

Circle Agent Marketplace Brings USDC Payments and Live Data Tools to Arc Blockchain

Key Takeaways: Circle Agent Marketplace now makes services available for AI agents on Arc. Agents have access to live web search, st | Bitget crypto news!

Bitget 1 week, 1 day ago

How OpenAI's AI Agents Chained Nine Zero-Days to Breach Hugging Face

[Share on X](https://twitter.com/intent/tweet?url=https%3A%2F%2Fstartupfortune.com%2Fhow-openais-ai-agents-chained-nine-zero-days-to-breach-hugging-face%2F&text=How%20OpenAI%27s%20AI%20Agents%20Chained%20Nine%20Zero-Days%20to%20Breach%20Hugging%20Face%20-%20Startup%20Fortune)[Share on LinkedIn](https://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Fstartupfortune.com%2Fhow-openais-ai-agents-chained-nine-zero-days-to-breach-hugging-face%2F)[Share on WhatsApp](https://wa.me/?text=How%20OpenAI%27s%20AI%20Agents%20Chained%20Nine%20Zero-Days%20to%20Breach%20Hugging%20Face%20-%20Startup%20Fortune%20https%3A%2F%2Fstartupfortune.com%2Fhow-openais-ai-agents-chained-nine-zero-days-to-breach-hugging-face%2F)[Share on Reddit](https://www.reddit.com/submit?url=https%3A%2F%2Fstartupfortune.com%2Fhow-openais-ai-agents-chained-nine-zero-days-to-breach-hugging-face%2F&title=How%20OpenAI%27s%20AI%20Agents%20Chained%20Nine%20Zero-Days%20to%20Breach%20Hugging%20Face%20-%20Startup%20Fortune)[Share via Email](mailto:?subject=How%20OpenAI%27s%20AI%20Agents%20Chained%20Nine%20Zero-Days%20to%20Breach%20Hugging%20Face%20-%20Startup%20Fortune&body=https%3A%2F%2Fstartupfortune.com%2Fhow-openais-ai-agents-chained-nine-zero-days-to-breach-hugging-face%2F) # How OpenAI's AI Agents Chained Nine Zero-Days to Breach Hugging Face [Share on X](https://twitter.com/intent/tweet?url=https%3A%2F%2Fstartupfortune.com%2Fhow-openais-ai-agents-chained-nine-zero-days-to-breach-hugging-face%2F&text=How+OpenAI%27s+AI+Agents+Chained+Nine+Zero-Days+to+Breach+Hugging+Face)[Share on LinkedIn](https://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Fstartupfortune.com%2Fhow-openais-ai-agents-chained-nine-zero-days-to-breach-hugging-face%2F)[Share on WhatsApp](https://wa.me/?text=How+OpenAI%27s+AI+Agents+Chained+Nine+Zero-Days+to+Breach+Hugging+Face+https%3A%2F%2Fstartupfortune.com%2Fhow-openais-ai-agents-chained-nine-zero-days-to-breach-hugging-face%2F)[Share on Reddit](https://www.reddit.com/sub

Startupfortune 1 week, 1 day ago