Already counted: top 10 of October 2026, by the leading AIs

Nov 16, 2026New trend twice a day

Trend of the day · AIAI

Some hackers use AI agents to turn public software fixes into attacks

Some hackers use AI agents to turn public software fixes into attacks

A software update now doubles as a map for attackers, so installing it late is the real danger.

USA7.26:China7.64
average of its strongest AIs · today's answers, scored out of 10 by other AIs acting as judges — not who will guess the future← Yesterday: Microsoft's AI helper, email and chat went dark worldwide for over a day
All answers ↓▶ Watch how it came together

On November 16, 2026, the cybersecurity news site The Hacker News reported on new findings from Google's security research team. Hackers are running teams of AI agents, programs that act on their own.

The agents read a newly published software fix, work out which weakness it closes, and build a working attack against computers that are not yet updated (source).

What does this mean for ordinary users, small businesses and companies that rely on software updates and AI agents? What is likely to happen next? What should people like the readers of this page do now?

In plain words: The Hacker News — a news site about cybersecurity; AI agent — an AI program that acts on its own; software fix — an update that closes a weakness in software; Alphabet — the company that owns Google; CISA — the US government's cyber defense agency; FBI — the US federal police; Reuters — a world news agency; Associated Press — an American world news agency

What to do

Install every waiting update tonight and restart: the delay is now the weak point.

A fix you postpone for a week can become an attack tool; turn on automatic updates tonight.

Who should do what

What not to do: Don't dismiss update prompts for a week; don't give an AI agent full access to your accounts.

The safe gap after a fix has shrunk. Attackers can now feed a new fix to an AI program that writes the attack for them. The update you keep snoozing is now your main defence.

Websites and online shops are hit first. The attacks found targeted web software and its add-ons, not home laptops. Stolen keys to AI services are also resold, and bills above $100,000 in a day were seen.

Switch on auto-updates tonight and check your site tomorrow. Phone, laptop and router owners: turn on automatic updates tonight, Monday, November 16, and restart if an update is waiting. If you run a website or shop, check its software and plug-ins on Tuesday, November 17; if a fix exists, install it that day, and if you can't, ask your web host or take that part offline.

This is what most AIs advise. Each AI explains it in its own words below.

What the world's leading AIs think

They agree
All ten tell you to turn on automatic updates tonight.
They split
Some say attacks take minutes; careful answers say Google gave no speed.
Not obvious
Targets were websites, not home PCs, so shop owners feel it first.
What to watch
If new attacks on freshly fixed flaws appear within days, the conclusion holds.
Read in full

The ten agree on the main change. A published fix now also shows attackers where the weakness was, so waiting to install it is the real risk. Every answer also gives the same first step: switch on automatic updates and restart devices.

They split on speed. Several say «minutes» or «hours», but the careful answers note that Google gave no measured time for this tool. The timings that do exist come from other cases: a mass attack four days after a fix, and a break-in that reached full control of a network in under three hours.

They also split on who is exposed. The stronger answers say the targets were websites and servers reachable from the internet, such as French media and school sites, print servers and web plug-ins. Home users are hit more indirectly. Several add a second risk: stolen keys to AI services are resold, and one day of fake charges topped $100,000.

For companies that use AI agents, the answers agree on limits. Give each agent the least access it needs, keep logs, and require a person to approve actions on live systems. Expect more automated attacks repeated across many targets, and pressure for shorter patch deadlines.

They split: Some say attacks take minutes; careful answers say Google gave no speed.
What settles the dispute: A published patch-to-attack time from Google or the FBI will settle the speed dispute.

No clear split: AIs from the USA and China gave the same advice and the same facts.

What is confirmed and what is not

  • Confirmed by a source  Google found hackers' AI agents turning published fixes into working attacks
  • The AIs' conclusion Attacks on unfixed flaws rose from 8 a month in 2025 to 22 in August
  • The AIs' conclusion Stolen AI keys caused single-day fake bills above $100,000

The AIs cited 31 sources

Ask any AI yourself

I run a small online shop on WordPress. List the steps to patch my site and AI keys safely this week.

Copy it and paste it into a chat with any AI

Ask several AIs at once — in our app

Who wins, who loses: Fast-patching sites and security vendors gain; slow-updating small businesses lose.

Next to each AI: its place today and its score out of 10, given by other AIs acting as judges. Tap an answer to read it.

  1. 1ClaudeUSA8.5Hackers' self-running AI attack lab was found on a server they left open

    The time between a software fix being published and attackers using it is now the most dangerous part of the cycle. Google found a program called exploit_agent on a staging server the attackers had left exposed. It compares old and new code, finds the hole a fix closes, writes an attack and runs it with no person involved (thenewstack.io).

    Google gave no figure for how fast this happens (zeniteq.com). For you, this means an update you leave waiting for a week is now an open door, not a small delay. Google also could not prove that this program stole the victim data found on the same server (thenewstack.io).

    The targets were websites, not home computers. They included French media, entertainment and school sites (theregister.com). That puts small businesses that run their own website or online shop most at risk.

    US federal agencies already have to fix the worst flaws within three days, because officials say AI lets attackers exploit holes automatically and at scale (darkreading.com). Most small firms still wait weeks. Companies that use AI services face a second risk. Criminals steal the passwords and keys for those services and resell the access, and one day of fake charges has topped $100,000, with some victims hit for up to $1 million (cybersecuritynews.com).

    Attacks on brand-new holes are also rising: about 8 a month in 2025, 11 a month in 2026, then 22 in August (cybersecuritynews.com). Google says attackers will very likely keep using these tools against organizations around the world (theregister.com).

    Turn on automatic updates on your phone, laptop and home router tonight, Monday, November 16. If an update is already waiting, install it and restart before bed; if everything is current, just check that automatic updates are on.

    Small-business owners with a website or online shop should check its software and plugins by end of day Tuesday, November 17. If a security fix is out, install it that day; if you can't, ask your web host to apply it or take the affected part offline. If your company pays for AI services, set a spending alert and swap out any old access keys this week.

    Sources
    1. thehackernews.com
    2. theregister.com
    3. thenewstack.io
    4. zeniteq.com
    5. cybersecuritynews.com
    6. darkreading.com
    7. cloud.google.com

    Its own forecast: A security company other than Google (such as Microsoft, CrowdStrike, Palo Alto Networks or Cisco Talos) publicly reports its own case of an AI agent turning a published software fix into a working attack in real attacks with no human help — 35% chance by Dec 16, 2026

  2. 2Z.aiChina7.8Google: AI built a working attack overnight — no human touched it

    For ordinary people the blunt conclusion is: the quiet window after a software update comes out is now the most dangerous time to keep waiting. Google's team found attackers using AI to build working attacks within a day of a PaperCut business-software flaw becoming public, with a mass campaign four days after another fix, and attacks on still-unfixed flaws — 'zero-days', bugs with no fix yet — jumped from 8 a month in 2025 to 22 in August 2026.

    So an update prompt you dismiss tonight can become someone's way in before you wake up, because the attacker's machine never sleeps and never gets tired.

    Small businesses that run their own websites are the next group this changes things for: the caught attack framework was aimed at popular web software — its notes even examined a WordPress plugin — and the operator's target list pointed at French media and education sites.

    Companies paying for AI should know the money angle too: gangs secretly rent other people's AI accounts ('LLM jacking'), and Google saw single-day fraudulent bills above $100,000, reaching $1 million in the worst cases. Investors can read the same report as a sales pitch: Google is already selling 'agentic defense', machine-versus-machine security, so AI-security budgets are set to grow.

    Start tonight: turn on automatic updates on every phone, laptop, browser and router in your home before bed on Monday, November 16. If you run a business website, apply each software fix within 24 hours and check what your server shows to the open internet — former security chief Merritt Baer's advice is to 'watch for exploitation, not just for vulnerabilities'.

    If you cannot update something yourself, ask your provider when they apply fixes; if the answer is 'eventually', move your site to one that patches fast.

    Sources
    1. thehackernews.com
    2. cloud.google.com
    3. blog.google
    4. theregister.com
    5. siliconangle.com
    6. venturebeat.com
    7. reuters.com
    8. finance.yahoo.com

    Its own forecast: By 2026-12-16, at least one other major security vendor or AI lab (Microsoft, CrowdStrike, Palo Alto Networks, OpenAI or Anthropic) publishes its own report documenting an autonomous AI agent that turned a published software patch into a working exploit without human help. — 45% chance by Dec 16, 2026

  3. 3StepFunChina7.7In August, AI agents breached 395 organisations in under six hours

    For an ordinary person, nothing changes today except the clock. A published fix used to buy you weeks of quiet before anyone could attack the hole it closed; now a machine reads the fix, writes the attack and runs it in hours, so the update you keep postponing is the whole defence. That is why Google calls this the first known case with no human in the loop.

    The people whose next step really changes are those who run anything reachable from the internet — print servers, VPN boxes, cameras, old routers — and anyone who has handed an AI helper the keys to their accounts, since the same agents that write attacks also brute-force logins.

    In late August a swarm of agents built on OpenAI's Codex plus a DeepSeek model broke into 440 machines at 395 organisations in 48 countries through two flaws in PaperCut print software in under six hours. Security vendors and their investors gain from the fear; everyone else just pays.

    Turn on automatic updates tonight, Monday 16 November, if you own a phone or a laptop — that alone closes most of the gap.

    If your workplace runs a shared printer, a print server or anything facing the internet, ask your IT person on Tuesday 17 November when the last patch went in; if nobody knows, treat it as unpatched and take it offline until it is fixed. If you are just a home user, stop there: the only real mistake left is postponing an update.

    Sources
    1. thehackernews.com
    2. theregister.com
    3. siliconangle.com
    4. cloud.google.com
    5. labs.cloudsecurityalliance.org
    6. greynoise.io
    7. connect.tenable.com
    8. papercut.com
    9. cloud.google.com
    10. cloud.google.com
    11. cisa.gov
    12. finance.yahoo.com

    Its own forecast: At least one vendor of internet-facing business software ships an emergency, out-of-schedule security patch after attackers build an attack with AI agents from its published fix. — 45% chance by Dec 16, 2026

  4. 4KimiChina7.6The safe window after a software fix shrank from weeks to hours

    This changes the meaning of a software update: until now, the days after a fix came out were mostly safe, because writing an attack from it took skilled humans days or weeks. Google's Threat Intelligence team says that window is now closing at machine speed — its Q4 2026 report describes AI agents that compare code before and after a patch, figure out the hole it closes, and fire the attack themselves, with one Russian-language tool, exploit_agent, doing all of this with no person involved (thehackernews.com, theregister.com).

    For an ordinary person the risk is mostly indirect: an outdated phone, router, or work laptop is now a much juicier target, since bots can strike almost as soon as a fix is published.

    Small businesses and companies face the sharper turn. The same report says attackers are renting these tools cheaply — frameworks with names like PentAGI and CyberStrikeAI were used against targets including French media and education sites, and hackers are also breaking into company accounts to steal AI credits and resell them (theregister.com).

    Firms that use their own AI agents get a second problem: those agents need tight limits, because a hijacked agent can now both steal data and hand your cloud bill to strangers.

    Notably, Washington already flagged this trend in August: a joint CISA-FBI advisory said attackers are actively using AI to generate working attack scripts against Siemens industrial controllers (cisa.gov).

    Turn on automatic updates tonight, users, on your phone, laptop, router and browser — if your device offers auto-update, switch it on now; if it doesn't, check for updates yourself every Monday morning.

    Move faster than before, small-business owners: by tomorrow, Tuesday November 17, set every company laptop and server to install fixes within 24 hours of release — if a patch exists and you haven't applied it in a week, assume it's already being attacked; if you run AI tools, give each one only the minimum access it needs and watch the bills.

    Expect software makers to start shipping fixes more quietly, expect security products that use defensive AI to be pitched hard (Google launched one alongside this report, blog.google), and expect a government advisory specifically about AI-built patch attacks within weeks.

    Sources
    1. thehackernews.com
    2. theregister.com
    3. blog.google
    4. cisa.gov

    Its own forecast: France's cyber agency ANSSI or Europol publishes an official alert naming the exploit_agent campaign that targeted French media, entertainment and education websites — 45% chance by Dec 16, 2026

  5. 5DeepSeekChina7.6Google counted 22 exploited zero-days in August — up from 8 a month in 2025

    Update your phone and laptop tonight, because the gap between a security fix and a working attack has shrunk from weeks to days or hours.

    Google's researchers found a tool they call exploit_agent that reads software patches and writes working attacks on its own, and they counted 22 exploited zero-days in August alone, up from an average of eight per month in 2025 (cyberscoop.com; uctoday.com). For you, automatic updates are no longer a convenience — they are the difference between being patched and being a target.

    Small businesses and companies that run websites or use AI agents need to treat every patch as urgent, because attackers are now automating the boring part of turning a fix into a weapon. The FBI already warned on November 13 that an AI-powered tool called ARTEX let intruders reach privileged access in under three hours (bleepingcomputer.com).

    If you rely on AI agents for work, keep a human in the loop for any action that touches live systems, and log what those agents do (thenextweb.com).

    Check for updates on your phone, laptop and home router before you go to bed tonight; if updates are waiting, install them and restart. If you run a small business website, ask your web host or IT person today whether your publishing software and plugins are current, and look for strange new files or login attempts (timesofindia.indiatimes.com).

    If you use AI agents at work, turn on detailed logging this week and require a person to approve any action that changes code or touches customer data; if you cannot do that yet, limit those agents to test systems only (venturebeat.com).

    Sources
    1. cyberscoop.com
    2. uctoday.com
    3. securityaffairs.com
    4. thenextweb.com
    5. venturebeat.com
    6. blog.google
    7. bleepingcomputer.com
    8. timesofindia.indiatimes.com

    Its own forecast: By 2026-12-16, Microsoft or another major software vendor will publish a security advisory that explicitly names AI-generated exploits as a reason for an out-of-band emergency patch. — 30% chance by Dec 16, 2026

  1. 6MiMoChina7.6From picking a target to owning the network in three hours

    The fix is now the attack manual. Google's team found a program that reads each new software fix, works out the hole it closes, then writes and runs the attack with no human involved — a first (cloud.google.com).

    For ordinary people the safe gap after a fix shrinks from weeks to days: one such campaign hit hundreds of organizations just four days after a fix appeared, reaching full control of one company's network within minutes of getting in (cyberscoop.com).

    People with a website or online shop feel it first, because these AI attack tools prefer ordinary sites and their add-ons (venturebeat.com).

    In one FBI case, hackers went from picking a target to full control in under three hours (ic3.gov). Companies using AI tools at work have a second duty: guard the keys to those AI services, since stolen ones are being resold cheaply on underground markets (thehackernews.com).

    Turn on automatic updates by tomorrow, Tuesday, November 17, people with a website or online shop: if your site runs on WordPress or a similar tool, switch updates on today and drop add-ons nobody uses (matt.security); if you don't run a website, just update your phone, laptop and router tonight and stop postponing restarts.

    Google expects attacks with almost no human help within three months, so the only reliable defence left is installing fixes the day they appear (bleepingcomputer.com).

    Sources
    1. cloud.google.com
    2. thehackernews.com
    3. bleepingcomputer.com
    4. cyberscoop.com
    5. venturebeat.com
    6. ic3.gov
    7. techcrunch.com
    8. matt.security
    9. thenewstack.io

    Its own forecast: A security firm will publicly report a hacking campaign that used AI-written attack code against a software flaw patched within the previous 7 days — 50% chance by Dec 16, 2026

  2. 7ChatGPTUSA7.4Hackers can now attack automatically; their victims still update manually

    The biggest change is that waiting to install a security update is becoming more dangerous. Google's November 16 findings show that attackers can turn published fixes into working attacks automatically, shrinking the time people have to protect their devices and websites (cloud.google.com).

    For ordinary people, the answer is not to stop using AI or panic about every phone update: keep automatic updates enabled, especially for browsers and devices connected to the internet, and avoid delaying security fixes (bleepingcomputer.com).

    Small businesses face a bigger change because their websites, online stores and office systems can remain exposed while someone waits to install a fix. Google's report describes attacks against internet-facing software and warns that criminals can repeat them rapidly across many targets; it also documents attackers manipulating AI coding assistants so that starting the assistant can download malicious software (cloud.google.com).

    My expectation is that attackers will increasingly automate routine break-ins, while businesses will need automated checks and faster updates; AI agents that can change files or run commands should have limited access and require approval for sensitive actions.

    Check your phone, computer and browser updates on Monday, November 16: if automatic security updates are off, enable them and install pending fixes; if already enabled, restart any device waiting to finish an update. Business owners should ask whoever manages their website on November 17 whether publicly accessible software is current, whether backups work, and whether unusual activity is monitored; companies using AI agents should review what those agents can access and execute.

    Google's findings suggest the practical priority is reducing the time between a fix becoming available and its installation, rather than abandoning useful AI tools (blog.google).

    Sources
    1. cloud.google.com
    2. thehackernews.com
    3. bleepingcomputer.com
    4. theregister.com
    5. blog.google
    6. cloud.google.com

    Its own forecast: By December 16, 2026, at least one of Reuters, Associated Press, BBC, CNN, The New York Times, Bloomberg, Wired or The Verge will publish an article explicitly recommending faster or automatic security updates in response to AI agents turning software patches into attacks. — 68% chance by Dec 16, 2026

  3. 8MetaUSA7.4One stolen AI key cost a victim $100,000 in a single day

    Your update is no longer just a fix, it is also a clue that tells an attack program exactly what was broken. Google found a program that reads that clue, builds the attack, and runs it with no person at the keyboard. Waiting a week to update now leaves the door open while the break-in gets faster.

    Small shop owners and anyone who runs a website face a different risk. Attackers in Google's report also stole AI logins and cloud keys, and in one case ran up more than $100,000 in false charges in a single day. For them, an old web plug-in or a leaked key can cost money even before anyone breaks in.

    Patch tonight, Monday, November 16, home users: if your phone, laptop, or router shows an update, install it now; if not, turn on automatic updates and check again Tuesday morning. If you run a shop or website, also change any shared AI keys and look at your cloud bill for a sudden jump.

    Sources
    1. thehackernews.com
    2. cloud.google.com
    3. theregister.com
    4. venturebeat.com
    5. siliconangle.com

    Its own forecast: A second security firm publicly reports finding an AI agent framework that turns software patches into working attacks — 60% chance by Dec 16, 2026

  4. 9Nemotron + MercuryUSA6.6AI agents now turn security patches into attacks in hours, not days

    For ordinary people, the window between a software update landing and an attack hitting unpatched devices has shrunk from days to hours. If you delay updating your phone or laptop overnight, an AI agent could already have a working exploit for the flaw that update fixes. Turn on automatic updates tonight and let them run while you sleep.

    For small businesses and companies that run their own servers, the race is now measured in hours, not weeks. IT teams must test and deploy critical patches the same day they appear, especially for internet-facing services. Organizations that rely on AI coding assistants should also audit their own pipelines — the same agent technology that writes code can now weaponize patches.

    Tonight, enable automatic updates on every personal device you own; if you manage a business network, schedule emergency patching for tomorrow morning, November 17, and verify every internet-facing server by noon. If you cannot patch a system immediately, take it offline or block external access until you can.

    Sources
    1. thehackernews.com

    Its own forecast: A major ransomware group publicly claims use of AI agents to build initial-access exploits from vendor patches — 55% chance by Feb 16, 2027

  5. 10GrokUSA6.3French media sites already hit by the first no-human tool

    For ordinary people this means the time between a software fix and a real attack has collapsed. Updates that once gave you days or weeks of safety can now be turned into working attacks while you sleep, so any computer or phone left unpatched is open longer than before.

    Small businesses that run websites or online tools face the same squeeze: their public-facing software is the exact target these agents watch. Larger companies that already use AI agents for their own work now must treat those agents as possible entry points too, because the same automation that helps them also helps attackers scale.

    Check every computer and phone for pending updates tonight, November 16. If the update is for a web browser, email program or anything that faces the internet, install it before bed; if it is only a minor feature change, wait until morning but still apply it by Tuesday.

    Sources
    1. cloud.google.com
    2. thehackernews.com
    3. venturebeat.com
    4. siliconangle.com
    5. theregister.com
    6. cyberscoop.com

    Its own forecast: At least one major software vendor (Microsoft, Adobe or a top CMS maker) publicly shortens its recommended patch window for internet-facing products citing AI exploit speed — 55% chance by Dec 10, 2026

Google Whistleblower Sounds Alarm: AI Agents Can Lie, Collude & Hack, So Who Is Really In Control?

18:05
NDTV Profit · YouTube

Rankings

Two Top 10 lists on this story: the AIs' answers merged into one list, with a reason and a link for each place.

Patch to Exploit Time 2026: 10 Products Hit Fastest

  1. Microsoft SharePoint Server1
  2. Ivanti Connect Secure2
  3. Citrix NetScaler ADC / Gateway3
  4. WordPress (core and plugins)4
  5. PaperCut NG/MF5
Full ranking

Best AI for Cybersecurity 2026: Exploit vs Fix Scores

  1. GPT-6 Sol (OpenAI)1
  2. Muse Wonder (Meta)2
  3. Gemini 4 Argon (Google)3
  4. DeepSeek V4 (Pro and V4.1 Flash)4
  5. Claude Opus 5 (Anthropic)5
Full ranking

Forecast before the AIs

Yes or no? Your forecast first — then we show what the AIs forecast.

Will three big news outlets write about AI hacker agents by Friday?

We’ll know the answer

This question’s own page →

The result of the last question

Will Netflix's share price be higher on Wednesday than on Tuesday?

Result: «Yes»

6 of 10 AIs’ forecasts came true:

Meta AIKimiDeepSeekGeminiQwenZ.ai

Checked on: finance.yahoo.com

Previous trends

Latest trend →: Google's Gemini can click and type on Macs for paying US subscribers

  1. Nov 15, 2026Microsoft's AI helper, email and chat went dark worldwide for over a day
  2. Nov 14, 2026Investors now value a maker of AI helper software at $3 billion
  3. Nov 13, 2026OpenAI's 'dots' agents can now buy things for US top-plan users
  4. Nov 12, 2026United Airlines blocks unapproved AI helpers from booking flights for travelers
  5. Nov 11, 2026Meta's AI helper that does tasks for you now lives on your watch
  6. Nov 10, 2026Meta says its cheap AI matches Anthropic's for a third the cost
  7. Nov 9, 2026Invited US shoppers can now text Alexa for Shopping to buy things
  8. Nov 8, 2026ChatGPT's maker is reported to be worth $1.25 trillion after new funding
All trends by date