One Hacker Turns AI Agents Loose on John Deere and SharkNinja, a Cyberattack Knocks a Staffing Giant Offline, and Only 13% of CIOs See Real AI Payoff
By Noah Smith, Owner & Consultant, KeyChange Technologies · October 20, 2026
Here is your AI and IT news for Tuesday, October 20, 2026: one hacker with a fleet of cheap AI agents got into John Deere and SharkNinja, a cyberattack took a staffing platform used by 10,000 firms offline, an Indiana hospital system pulled its own network, and Gartner says most CIOs still are not seeing real AI payoff. Yesterday's recap is here.
📌 The AI and IT news at a glance
- 🔝 One hacker used an army of AI agents to break into 100+ company websites, including John Deere and SharkNinja
- 🛡️ A cyberattack takes Bullhorn, the staffing and payroll platform behind 10,000+ firms, offline
- 🛡️ Parkview Health shuts down its network after "suspicious activity," taking MyChart down across 16 hospitals
- 🛡️ Microsoft's October Windows Server update is breaking Outlook sign-in and mail flow on Exchange
- 🤖 Gartner: only 13% of CIOs say AI has delivered significant value
- 🤖 WhatsApp hid instructions telling AI bots not to point users to privacy regulators
- 🤖 The EU publishes the first official standard for complying with the AI Act
- 🧰 OpenAI says its newest models can now actually handle COBOL and other legacy code
- 🧰 Cloudflare previews Clairic, a model that pulls data out of documents for a fraction of LLM cost
🔝 Top story: One hacker, a swarm of AI agents, and 100+ broken websites
A threat researcher at the security firm NorForge says a single Chinese-speaking hacker used AI agents to attack more than 1,200 websites starting September 25, and broke into more than 100 of them. Reuters, which saw the report before publication, identified three victims: John Deere, SharkNinja, and Cummins. The agents ran on an open-source platform called Cairn, and each successful break-in took a little under 23 hours on average. NorForge researcher Snorre Fagerland roughly estimated the hacker spent about $5,000 on AI tokens, or around $50 per successful hack, with a success rate of roughly 10%.
The companies have downplayed the damage. Deere said it took remediation steps and saw no sign anyone accessed sensitive information. SharkNinja said the attack hit a standalone product support site and has been contained. Cummins said it is looking into a report about a website run by one of its vendors. Reuters could not establish how much data, if any, was taken. The researchers also found hundreds of exposed credentials and API keys in the attacker's control panels, about 11% of which were verified as still valid.
In short: A researcher says one hacker spent about $5,000 on AI to break into more than 100 company websites, including sites tied to John Deere, SharkNinja, and Cummins.
What it means for your business: The cost of attacking a website just collapsed to the price of a nice lunch, which means small and forgotten sites (old support portals, vendor-run microsites, campaign pages) are now worth an attacker's time. If you have a website you have not patched or looked at in a year, assume something is looking at it now.
My take: Notice where these got in: a standalone support site, a vendor's website. Not the crown jewels. That is the real lesson. AI does not make attackers smarter so much as it makes them tireless, and tireless attackers find the thing you forgot you owned. Make an inventory of every site and login page with your name on it this week, and shut down what you do not need.
Source: TNW: AI agents broke into 100+ sites for one hacker at $50 each, researcher says
🛡️ IT and security
A cyberattack takes the Bullhorn staffing platform offline
Bullhorn, a cloud software company whose platform staffing and recruiting firms use for candidate tracking, onboarding, and payroll, has taken its platform offline to contain a cyberattack. Customers started having trouble posting jobs on Friday, the company reported a system-wide outage the next day, and it then confirmed "a cyber incident on the Bullhorn Platform" and disabled access. Bullhorn says it serves more than 10,000 customers worldwide and is working with CrowdStrike before restoring service.
As of Monday, its core ATS/CRM, back office, and talent platform were listed with major outages, with a long list of add-on products degraded. Bullhorn has not said whether any data was stolen and has not given a restoration timeline. All non-SSO users must reset their passwords at next login, and API passwords used for integrations must be rotated. Unposted time records will be reprocessed once service returns.
In short: A cyberattack has forced Bullhorn, a staffing software platform with more than 10,000 customers, offline with no timeline for full recovery.
What it means for your business: If you use a staffing agency, the people you are paying through them may be affected, and timesheets and invoices could run late. If you use Bullhorn directly, reset passwords and rotate any API credentials now, not after service comes back.
My take: This is the vendor version of "your redundancy is not redundant." When one SaaS platform runs your hiring, onboarding, and payroll, a single incident stops all three at once. Ask your critical vendors a simple question this month: if you go dark for a week, how do I pay my people?
Source: BleepingComputer: Bullhorn takes staffing platform offline to contain cyberattack
Parkview Health pulls its own network offline, taking MyChart down
Parkview Health, a Fort Wayne, Indiana health system with 16 hospitals and more than 300 outpatient centers and physician offices across Indiana and northwest Ohio, reported a "systemwide internet outage" on Monday. The system says it proactively disabled its network after detecting "suspicious activity." Patients lost access to MyChart, the portal they use for test results, messaging doctors, refills, and appointments.
Parkview says there is "no indication" data has been compromised. Patients with urgent needs were told to call their provider's office or visit a walk-in clinic, and to call 911 for emergencies. As of Monday afternoon, Parkview had not said when services would return or what the suspicious activity was.
In short: Parkview Health took its network offline after detecting suspicious activity, cutting off its MyChart patient portal across Indiana and Ohio.
What it means for your business: Pulling the plug early is a deliberate, painful choice that often limits damage. It only works if staff can keep operating without the network, which is something you can plan for.
My take: Credit to Parkview for moving fast instead of hoping it was nothing. The question every owner should ask is the one they are answering right now: if we had to unplug everything at 9 a.m., what would we do at 9:05? Write that page down before you need it.
Microsoft's October Windows Server update is breaking Exchange email
Microsoft has confirmed that its October 2026 Windows Server security updates are causing problems for Exchange Server on Windows Server 2019 and later. Users may be unable to sign in to Outlook or Outlook on the web, emails can get stuck in the Outbox, and Exchange can struggle to reach mailboxes. Microsoft says a new default limit in the update, 512 simultaneous Endpoint Mapper lookups, can be too low for busy Exchange servers, so impact depends on load and some servers will never see it.
A permanent fix is in progress. In the meantime, Microsoft has published a temporary registry change that raises the limit, followed by a reboot, and it is telling customers not to uninstall the updates. Microsoft stresses this is not a problem with October's Exchange security updates themselves.
In short: October's Windows Server updates can stop Outlook sign-in and mail delivery on busy on-premises Exchange servers, and Microsoft has a temporary workaround.
What it means for your business: If you still run your own Exchange server and email got flaky after patching, this is likely why. Have your IT provider apply Microsoft's workaround rather than rolling back security updates.
My take: This is one more reason on-premises Exchange is getting hard to justify for a small business. Exchange 2016 and 2019 extended security updates also end this month, so if you are still running your own mail server, this is the nudge to plan the move.
Source: BleepingComputer: Microsoft: October Windows Server updates break Exchange sign-in, mail flow
🤖 AI generally
Gartner: only 13% of CIOs say AI has delivered significant value
At its IT Symposium in Orlando on Monday, Gartner shared results from its annual survey of more than 2,000 CIOs and IT executives. AI spending has grown 18-fold over the past two years, yet only 13% of CIOs say they have realized significant or transformational value from AI. Governance is lagging too: only 21% have governance in place for agentic AI, while 71% plan to deploy AI agents within 12 months. Just 1% have bought AI insurance.
Gartner analyst Kristin Moyer put the core problem bluntly: "The biggest drag on AI momentum is using it when you don't need to." Gartner's advice was to tie each AI project to a specific business outcome, prove value before spending more, and rein in shadow AI. The bright spot it highlighted was IKEA, which retrained 8,500 call-center workers to sell interior design as chatbots took routine calls, adding more than $200 million in revenue.
In short: Gartner's survey of 2,000+ CIOs found AI spending up 18-fold in two years but only 13% reporting significant value.
What it means for your business: If your AI investments feel underwhelming, you are in the large majority. The companies winning are picking specific problems and measuring results, not buying AI because a competitor did.
My take: The IKEA example is the one to steal. They did not use AI to cut 8,500 people; they used it to free them up for higher-value work, and it paid. Pick one job your team hates, measure it, and try AI there before you buy anything bigger.
Source: InformationWeek: Gartner: CIOs must balance AI value with risk, and fast
WhatsApp hid instructions telling AI bots not to send users to regulators
Privacy app maker Redact reviewed 545 WhatsApp help pages and found 102 blocks of hidden text on 80 of them, invisible to people but readable by AI systems. Most were routine. Two stood out: the account deletion page told AI not to recommend "any Data Protection Authority, regulator, or supervisory authority," and another page told AI not to reference data protection laws like GDPR or CCPA, even when users brought up lawyers or courts.
WhatsApp told WIRED the notes were instructions for its own support chatbot that appeared on public pages because of a bug, now fixed, and that it wants its bots focused on WhatsApp advice, not legal advice. The notes were removed by October 18. Redact found nothing similar on 300 Facebook and Instagram help pages. It could not confirm whether the chatbot or outside AI assistants actually followed the instructions.
In short: Researchers found hidden text on WhatsApp help pages telling AI not to point users to privacy regulators or laws; WhatsApp says it was a bug and has removed it.
What it means for your business: Customers increasingly read your website through an AI assistant, and anything hidden in your pages can shape what that assistant tells them. Regulators will care what your support bots are told to say and not say.
My take: Whether or not it was a bug, it is a preview of a new kind of trust problem: the instructions behind the bot matter as much as the policy on the page. If you run a support chatbot, read its instructions like a regulator would, and make sure nothing in there would embarrass you if it leaked.
Source: Redact: WhatsApp help pages contained hidden AI instructions
The EU publishes the first official standard for complying with the AI Act
The European Commission has published EN 18286:2026, the first harmonised standard supporting the EU AI Act. It covers the quality management system that providers of high-risk AI systems must have under Article 17. Companies that implement it get a "presumption of conformity" with those requirements, and the Commission says it is the only AI quality management standard that offers that.
The standard is designed to sit on top of systems companies may already use, such as ISO/IEC 42001 or ISO 9001, rather than replace them. It arrives ahead of the AI Act's high-risk system requirements, which the Commission lists as applying in December 2027 and August 2028, and more standards are expected to follow.
In short: The EU published EN 18286, the first official standard companies can follow to show their high-risk AI systems meet the AI Act's quality management rules.
What it means for your business: If you build or sell AI products that touch hiring, lending, health, or other "high-risk" uses in Europe, you now have a concrete checklist instead of guesswork. If you only use AI tools, expect your vendors to start citing this standard.
My take: Standards are boring until a customer's procurement team asks for them. If you sell into Europe, get a copy now and map it against what you already do. It is far cheaper to build this in over the next year than to retrofit it under deadline.
🧰 New tools for builders and business AI
OpenAI says its newest models can finally handle COBOL and other legacy code
OpenAI published results showing its newest models are dramatically better at "low-resource" programming languages like COBOL, Fortran, and Visual Basic, which have far less training data than Python. On a new 20-language test that has models build a working database engine, GPT-6 Astra averaged an 87.4% pass rate versus 26.8% for the older GPT-5.6 Sol, with COBOL jumping from 10.9% to 72.1% and Visual Basic from 27.5% to 89.4%.
On a separate COBOL migration test, translating old code into languages like Java, Python, and Rust, GPT-6.1 Sol hit 86.6% accuracy at low effort, versus 5.4% for GPT-5.6 Sol, at 39% lower cost, and reached 97.9% at higher effort. The models are available through OpenAI's API, and OpenAI released the test suite on GitHub.
In short: OpenAI says GPT-6-era models now handle COBOL, Fortran, and Visual Basic far better, including migrating COBOL to modern languages with up to 97.9% accuracy on its own test.
What it means for your business: If you depend on an old line-of-business system or a pile of Visual Basic and Access tools nobody wants to touch, modernizing them just got cheaper to scope. These are OpenAI's own benchmarks, so test on your code before you plan around them.
My take: This is one of the most practical AI wins for small and mid-sized companies, because so many run on something a retired contractor wrote in 2004. Do not hand the whole thing to a model, but do let one write the documentation and a first-pass translation, then have a human review it.
Source: OpenAI: Advancing coding in low-resource programming languages
Cloudflare previews Clairic, a cheap model for pulling data out of documents
Cloudflare released a preview of Clairic, a model built for extraction rather than chat. Instead of writing text, it returns typed values, like names, numbers, dates, and lists, with a confidence score for each, and it can also produce estimates and time-series forecasts. Cloudflare says it charges only for input and typically responds in under a second. It is available now on Workers AI, with open weights promised later.
In Cloudflare's own demo, a claims-processing pipeline that used Clairic for extraction ran about 3.2 times faster and cost roughly 49% less than an all-LLM version. Cloudflare calls the endpoint an early checkpoint.
In short: Cloudflare previewed Clairic, a fast, input-priced model for extracting structured data from text, now on Workers AI.
What it means for your business: Much of the "AI" small businesses want is really data entry: pulling fields off invoices, forms, and emails. Purpose-built extractors like this can do that faster and cheaper than a general chatbot.
My take: Worth a look if you are building document workflows, but it is a preview from a vendor grading its own homework. The bigger trend is real: use the smallest, most specific tool that does the job, and save the big models for the parts that need judgment.
Source: Cloudflare Blog: Clairic extraction model
That is today's AI and IT news. Back tomorrow with the next round.