cd /news/ai-safety/ai-news-report-november-16-fbi-ai-ha… · home › topics › ai-safety › article
[ARTICLE · art-170704] src=theainewsreport.com ↗ pub= topic=ai-safety verified=true sentiment=↓ negative

AI News Report, November 16: FBI: AI HACKING AGENT ARTEX REACHED 2,250 TENANTS VIA CONNECTWISE BACKUP

The FBI issued a FLASH alert on ARTEX, an open-source framework that lets AI agents carry out intrusions with little human help, with some attacks reaching privileged access in under three hours and one campaign through ConnectWise Cloud Backup enabling access to more than 2,250 downstream tenants. Amazon, Cloudflare, Cisco Talos, Google and Barracuda published related findings in the same period, and the FBI said the common entry point was a forgotten credential, urging organizations to revoke unused vendor keys, limit pipeline tokens and review Microsoft 365 app consent by application ID rather than display name.

by read3 min views1 publishedNov 16, 2026
AI News Report, November 16: FBI: AI HACKING AGENT ARTEX REACHED 2,250 TENANTS VIA CONNECTWISE BACKUP
Image: Theainewsreport (auto-discovered)

Google's threat team found a framework that compared code before and after patches, wrote exploits and ran them. Patch internet-facing apps and WordPress plugins in days, not weeks.

One click on a malicious page could add a trusted filter that injects script anywhere. Make sure managed browsers run 1.76.x; 1.75.4 was the first fixed build.

Both scored 150 of 150 in every test group, but Fable cost an estimated $8.98 against $2.29. The author says start with Opus 5.5 and move up only when it falls short.

It parses HTML, runs page JavaScript and renders screenshots with no browser container, and Puppeteer can connect to it. Point one of your scripts at it to see what renders.

On FrontierML a human gains 66.1 points with a $200 budget, Opus 5.5 gains 58.8 and Kimi K3.1 Pro gains 29.6. A yardstick for research agents worth bookmarking.

He argues AI could match the best humans at AI research within a couple of years, maybe months. Read it for the evidence he lines up, then judge the forecast yourself.

CVE-2026-99288 lets a remote sender panic the kernel, taking down every container on the host. Patch the kernel; turning off SYN cookies in each namespace is a stopgap.

Huntress found 2,408 guest sessions with names like KUCOIN-SRV since November 3. Update technician clients to 26.6.5 or later, and do not join a session you cannot tie to a customer.

SANS saw attempts to chain two bugs into code execution on the LiteLLM proxy with no login. Upgrade past 1.83.7, block /mcp-rest/test/ from outside and rotate provider keys.

Cursor says at most 5,000 users may have received the bad cached versions, which stayed up about 11 hours. It has moved all inference traffic off Fireworks.

ZDNET reports GPT 6 Sol found the bug in 13 minutes in September, after a February test spent 25 hours and $350 and failed. Run AI review on your public apps first.

November's updates broke activation for shops that use SLMGR, and the fix came Friday. If you script activation, test the new KB before your next patch cycle.

The BriefThe FBI issued a FLASH alert on ARTEX, an open-source framework that lets AI agents carry out break-ins with little human help. Some intrusions went from first target to privileged access in under three hours, and one campaign through ConnectWise Cloud Backup enabled access to more than 2,250 downstream tenants. Amazon, Cloudflare, Cisco Talos, Google and Barracuda published their own findings in the same days, and the common way in was a credential someone forgot, so start by revoking vendor keys you no longer need, limiting pipeline tokens, and reviewing Microsoft 365 app consent by application ID.

Level UpReview which apps hold consent in your Microsoft 365 tenants. In the Entra admin center, go to Entra ID, Enterprise apps, All applications, pick an app, open Permissions and check the Admin consent tab. The FBI says to search by application ID, not by display name, since a name can be copied. →Microsoft Learn: Review permissions granted to enterprise applications

Caught Up?

Mon 11/9: Outlook.com bounced mail from some senders with error 550 5.7.23.

Tue 11/10: N-able took hosted N-central offline after more than 30 zero-days went public.

Wed 11/11: A Microsoft Outlook zero-day leaked Windows login hashes from one opened file.

Thu 11/12: Apache fixed an HTTP/2 bug that let attackers run code; 2.4.71 is the fix.

Fri 11/13: A Microsoft 365 outage ran past 20 hours; Microsoft now blames a code change from July.

── more in #ai-safety 4 stories · sorted by recency
── more on @fbi 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/ai-news-report-novem…] indexed:0 read:3min 2026-11-16 · —