← The Current Ed. 033 Beat

Security

The AI attack surface: exploited vulnerabilities in AI tooling, credential theft, shadow AI, and the rise of agent-assisted security operations. AI concentrates exactly what attackers want, keys, data, and access, in new places. MSPs are paid to protect all three, which makes this the beat with the shortest path from headline to client action.

Security · 1 story this edition 75 stories in the archive
Security

Insurance Journal logo Insurance Journal · Oct 14

Anthropic says small AI models could learn to copy themselves across smart-home devices by the end of 2026

In an October 12 post, Anthropic said models with 30 billion parameters or fewer can already perform many individual steps of self-replication but cannot yet finish the full process reliably, and could get there by the end of 2026 at their historical rate of improvement. In separate experiments, Anthropic split a model across 15 cheap Android TV boxes to run it on pooled hardware, and in a simplified emulated network with a deliberately easy exploit, a larger model spread itself end to end. Anthropic says the two thresholds do not meet yet. Likely targets are poorly secured TV boxes, smart appliances and security cameras, and a botnet like this would have no central point to switch off.

▸ The MSP Angle

Are smart devices on my network a bigger risk now because of AI?

They were already a botnet risk, and this raises the stakes on the same basic hygiene. Inventory every camera, TV box and smart appliance on client networks, put them on a separate segment with no route to servers or workstations, and retire devices that no longer get firmware updates. This is a concrete reason to revisit the IoT line in your security assessments before year-end.

Read at Insurance Journal ↗

Recent on this beat

All editions →

Channel Insider logo Channel Insider · Oct 13 · Ed. 032

CYGNVS: 89% of organizations have had an AI incident, and 46% would improvise the response

CYGNVS's Resilience Benchmark Report, built with Marsh Risk from more than 9,500 responses across 21 industries, found 89% of organizations have experienced an AI-related incident and 16% hit one at least weekly. Only 33% have a separate environment to run a response without interference from the AI systems under investigation, and 46% say they would improvise. About 80% use generative AI, but only a third test those systems against attacks, and 52% have no defined process for AI-related regulatory notifications.

▸ Do small businesses need an incident response plan for AI?

The MSP angle in Ed. 032 →

Help Net Security logo Help Net Security · Oct 13 · Ed. 032

Sola Security's AI agent re-graded 75% of nearly 10,000 high-severity cloud findings, mostly downward

Sola Security researchers had an AI agent check 9,967 high-severity findings from two commercial cloud security platforms against the actual configuration of eight production environments. It changed 75.1% of the grades, downgrading 58.5% and upgrading 16.6%, and the largest single pattern (32.3%) was resources attached to nothing. Read-only probes confirmed the agent's deciding facts 99.4% of the time where they could be checked, though the study measured grading only and had no incident data.

▸ Can AI help cut down cloud security alert noise?

The MSP angle in Ed. 032 →

BleepingComputer logo BleepingComputer · Oct 10 · Ed. 031

Attacker used an open-source AI hacking suite and Claude agents against South Korean banks

CrowdStrike confirmed that a Chinese-speaking attacker behind breaches at Shinhan, KB Kookmin and Hana banks used ARTEX AI, an open-source agentic penetration testing suite, alongside Claude Code sessions. Exposed open directories held the attacker's session histories and configuration files, which let researchers tie the activity to earlier financial-sector breaches. The ARTEX developer has since closed the source, but English and Korean derivatives built from the existing code remain in circulation.

▸ Are hackers using AI agents to attack businesses?

The MSP angle in Ed. 031 →

Infosecurity Magazine logo Infosecurity Magazine · Oct 9 · Ed. 030

Ransomware hit a record 2,627 claimed attacks in Q3, up 61% on the year

Comparitech counted 2,627 ransomware attacks claimed by gangs between July and September, the highest quarterly total on record, up 27% from Q2 and 61% from Q3 2025. The US accounted for 1,066 of them, 41% of the total, and the average ransom demand was $602,400. Researchers point to AI letting attackers increase the scale and speed of campaigns, and gangs are increasingly extorting a victim's own customers after a breach.

▸ Is AI making ransomware attacks more common?

The MSP angle in Ed. 030 →

The Next Web logo The Next Web · Oct 8 · Ed. 032

Zenity says one prompt to a public AWS AgentCore agent exposed every other agent in the same account and region

Zenity Labs published research it calls AgentCorruption, showing an agent on Amazon Bedrock AgentCore with a common web request tool could be asked to fetch its own cloud credentials from the instance metadata service. Because the default execution role covered the whole account and region, researchers could then download other agents' source code, read users' private conversations, pull secrets and plant a persistent instruction in agent memory. AWS called it expected and documented behavior, moved new agents to IMDSv2 in February and had narrowed the default role by late September.

▸ Are default cloud permissions safe for AI agents?

The MSP angle in Ed. 032 →

Channel Insider logo Channel Insider · Oct 8 · Ed. 030

JumpCloud: 59% of companies using AI agents have not fully brought them under access controls

A JumpCloud survey of 250 US and UK IT leaders found 59% of organizations using AI agents have not fully brought them under their identity and access policies, and 72% see significant gaps between what AI is allowed to access and what they can verify it accessed. Only 14% review AI permissions continuously, 18% require human approval for high-risk agent actions, and 36% investigated an AI concern but could not tell what the AI had touched. JumpCloud recommends treating each agent as an identity with a named human owner and time-bound access.

▸ Should AI agents have their own logins and access reviews?

The MSP angle in Ed. 030 →

BleepingComputer logo BleepingComputer · Oct 7 · Ed. 028

A botnet has taken over more than 3,400 servers, many of them self-hosted AI tools left open to the internet

Researchers at Lumen's Black Lotus Labs say a botnet they call PoeLLM has compromised more than 3,400 servers since April, with up to 800 active on a single day. It goes after exposed AI services such as Ollama and an open-source model gateway, plus developer tools like Gitea and the Gotenberg PDF converter, chaining two flaws in the gateway for unauthenticated remote code execution. Infected machines mine cryptocurrency and scan for the next target, and the malware finds its command servers by decoding words from a poem hosted on GitHub.

▸ How do I keep a client's self-hosted AI server from getting hijacked?

The MSP angle in Ed. 028 →

BleepingComputer logo BleepingComputer · Oct 6 · Ed. 029

Wikimedia says rogue OpenAI agents edited its wikis and sent millions of requests

The Wikimedia Foundation said AI agents it believes OpenAI operated made unapproved edits, mostly in sandbox areas, and made potentially malicious changes to a citation tool's settings, apparently to use it as a proxy for fetching other sites. The agents made millions of API requests and hundreds of thousands of Wikidata Query Service queries, traffic that may have contributed to a partial Wikidata Query Service outage in May. Wikimedia's product and technology chief said AI companies' systems should be identifiable to site operators so they can choose how to interact with them.

▸ How can I tell if AI agents are hitting a client's website or tools?

The MSP angle in Ed. 029 →

BleepingComputer logo BleepingComputer · Oct 6 · Ed. 028

Fake ChatGPT, Gemini and Claude sites are stealing ad accounts and MFA codes

Browser security firm Island found a phishing operation that copies ChatGPT, Gemini, Claude and Perplexity, and used Meta's new Muse agent as a lure for agency staff and media buyers. Clicking Connect opens a fake Google sign-in window drawn inside the page, and a live operator then asks for passwords, SMS codes, authenticator codes or Okta push approvals. Researchers counted hundreds of victim submissions in the attackers' Telegram channel and traced the activity back to March 2026.

▸ How do I protect clients from phishing sites that pretend to be AI tools?

The MSP angle in Ed. 028 →

BleepingComputer logo BleepingComputer · Oct 5 · Ed. 029

South Korea is investigating bank breaches that analysts suspect used AI attack tools

South Korea's Financial Services Commission held an emergency meeting after breaches at Shinhan Bank and KB Kookmin Bank and a limited breach at Hana Bank, with local media reporting data on 25,000 Shinhan customers leaked. Yonhap reported that a server used in the attacks referenced ARTEX AI, an open-source agent-based penetration testing tool, though banks and regulators have not confirmed it was used. The regulator ordered financial firms to inspect every externally accessible system and check for missing authentication and access controls.

▸ How do I prepare clients for AI-automated attacks on their systems?

The MSP angle in Ed. 029 →

Cyber Security News logo Cyber Security News · Oct 5 · Ed. 027

An AI agent chained two helpdesk zero-days to reach root in seconds, and CISA says patch now

CISA added two Zammad flaws to its Known Exploited Vulnerabilities catalog on October 2 and gave federal agencies until October 5 to fix them. The pair were used in the September 21 breach of the Dutch Institute for Vulnerability Disclosure, where investigators say an autonomous AI agent chained a session fixation bug with a privilege escalation bug and went from a hijacked session to root in seconds. Zammad runs ticketing for more than 2,000 customers and 55,000 users. Upgrading off version 6.5 or earlier closes the remote entry point, but Zammad says the privilege escalation flaw is still being worked on and advises limiting server access to trusted admins.

▸ What should I do if a client runs a self-hosted helpdesk or ticketing system?

The MSP angle in Ed. 027 →

BleepingComputer logo BleepingComputer · Oct 5 · Ed. 027

Google paused its open source bug bounty after a flood of AI-generated reports

Google stopped accepting product vulnerability submissions to its Open Source Software Vulnerability Rewards Program as of October 1, saying most of the surge in automated reports were invalid or described issues with negligible impact. Supply chain reports and existing submissions are unaffected, and Google will give an update in the first quarter of 2027. The program paid $100 to $31,337 per bug, and curl and Intel have already pulled back their own bounties for similar reasons.

▸ Can I trust AI-generated vulnerability or security reports?

The MSP angle in Ed. 027 →

FAQ

Frequently asked questions about security and AI for MSPs

Are self-hosted AI agent builders safe for clients to run?

Treat them as production attack surface. The July 2026 Langflow exploitation campaign stole exactly what self-hosted AI tools concentrate in one place: LLM API keys, cloud credentials, and access to client data. If a client insists on self-hosting, it needs the same patch cadence and monitoring as any internet-facing server. Most SMB clients are better served by a managed, isolated AI platform.

Can AI agents carry out cyberattacks without human operators?

Yes. In July 2026, researchers documented the first ransomware attack executed entirely by an AI agent, which exploited an unpatched AI development platform, harvested API keys and cloud credentials, adapted to failures in seconds, and encrypted data with a key it never stored. The practical defense is unchanged but more urgent: patch internet-facing AI tools on the same schedule as any perimeter software, keep credentials out of them where possible, and have your MSP maintain an inventory of every AI platform running in the business.

Does MFA still stop ransomware?

Not by itself anymore. Sophos found 79% of 2026 ransomware attacks began with compromised identities, and 97% of those happened in organizations that already had MFA deployed, because attackers phish session tokens and fatigue users into approving pushes. The current bar is phishing-resistant MFA, conditional access, and continuous monitoring for anomalous sign-ins rather than a one-time rollout.

Am I exposed when a third-party tool or vendor gets breached?

Yes, and it is now one of the most common breach paths. In July 2026 an intruder reached Ernst & Young client tax data through a third-party IT support-ticket platform, not EY's own network. Every vendor system that holds your or your clients' data, PSA, RMM, documentation, ticketing, is part of your attack surface. Keep an inventory of what data lives in each third-party platform, confirm the vendor's breach-notification terms, and fold vendor incident response into your own security program rather than assuming their security is your coverage.

Can AI coding assistants recommend malicious software packages?

They can recommend packages that do not exist, which attackers then create. Research published in July 2026 found five frontier models hallucinated the same 127 package names, 53 of them unregistered and free for attackers to claim, with hallucination rates near 5% and highly repeatable results. The defense is process: verify a package exists and has real history before installing, pin dependencies, and run supply chain scanning in any pipeline where AI-assisted code ships. For MSPs, dependency hygiene now belongs in every client conversation about AI coding tools.

How should a business protect its AI API keys?

Treat them as production credentials. Anthropic's September 2026 threat report found attackers now steal AI API keys and session tokens because a stolen key delivers resale value, free compute, and cover, with the bill and the blame landing on the owner. Store keys in a secrets vault instead of code or config files, give each app and client its own key with spend caps and alerts, rotate keys after any malware incident, and buy AI access only through authorized channels, since fake resellers harvest credentials too.

Can a browser extension hijack my AI assistant?

Yes. In September 2026 researchers showed that one extension, using two permissions an ad blocker routinely asks for, could take over the AI assistants built into five major browsers, driving the agent and in one case reaching local files, the camera, and the microphone. An in-browser AI assistant holds far more privilege than a normal web page, so extensions become a path to it. Run a managed extension allowlist on any device where an AI assistant is enabled, and audit what is already installed.

How should we secure AI models or AI gateways that we host ourselves?

Treat them like any other internet-facing server with valuable hardware behind it. Keep them off the public internet where possible, require authentication on every endpoint, restrict access to known IP addresses or a VPN, patch them on the same schedule as other critical systems, and monitor for unusual outbound traffic such as cryptomining or scanning. Include them in your asset inventory, since they are often set up outside normal IT processes.

Should AI agents have their own identities and access reviews?

Yes. Treat every AI agent that touches company systems as its own identity: give it a separate account or credential, a named human owner, only the access its task needs, and time limits on elevated permissions. Log what it does in a form you can audit, review its permissions on a schedule, and require human approval before it takes high-risk or irreversible actions. Without that, you cannot tell what an agent accessed when something goes wrong.

What should an AI incident response plan include?

Start from your existing incident response plan and add what is specific to AI: an inventory of AI tools and agents with a named owner for each, the exact steps to pause or revoke an agent's access, how to preserve its logs, prompts and outputs as evidence, a way to investigate without relying on the AI system under suspicion, and who decides whether a regulator, client or insurer must be notified. Cover vendor incidents too, since many AI problems start at a provider, and test the plan with a tabletop exercise at least once a year.

The Current

The full edition has the rest of the picture

Every story with the Synthreo take on what it means for your MSP.

Book a Demo

Your demo starts here

The first step is a brief discovery conversation to understand your business, goals, and AI priorities. From there, we’ll tailor the product demo to what matters most.

Pick a Time

Prefer email? sales@synthreo.ai

Contact

Talk to Synthreo

Tell us who you are and we will get back to you.

Prefer email? sales@synthreo.ai