Weekly Cybersecurity Recap: 98 Zero-Days, Rogue AI Agents, $100M Crypto Theft and More

Cybersecurity threats do not always begin with suspicious emails or obvious system failures. Sometimes, attackers exploit software that organizations already trust, devices that appear secure, or AI tools that have been granted too much access.

This week’s cybersecurity developments highlight several major risks: 98 zero-day vulnerabilities demonstrated at a hacking competition, unauthorized activity involving AI agents, cryptocurrency theft linked to compromised hardware, and cyberattacks that disrupted refrigeration systems at hundreds of supermarkets.

The common lesson is clear: security depends not only on protecting systems from outsiders but also on verifying the software, devices, services and permissions that organizations rely on every day.

⚡ Threat of the Week: ConnectWise Suspends Cloud Backup

ConnectWise suspended its Cloud Backup service after investigating unauthorized access involving application identities and compromised Microsoft 365 tenants. The company also disabled seven application identities and revoked certificates associated with affected connections.

This incident highlights the risks associated with application identities and trusted cloud integrations. These connections can have legitimate access to business information, so unauthorized access to them can create a serious security problem.

Organizations should review application permissions, investigate unusual sign-in activity, rotate potentially exposed credentials and follow official vendor guidance during incident response.

Security takeaway: Cloud security requires monitoring both user accounts and application identities.

🔥 Top Cybersecurity Stories This Week

1. Pwn2Own Ireland Demonstrates 98 Zero-Day Vulnerabilities

The Pwn2Own Ireland 2026 competition saw security researchers demonstrate exploits against 98 previously unknown vulnerabilities, earning a combined $1,262,000 in rewards. The event included demonstrations involving Google Pixel 10 devices.

A zero-day vulnerability is a software or hardware security flaw for which an effective fix may not yet be available when exploitation begins. Such weaknesses can give attackers opportunities to compromise affected systems before defenders have deployed a patch.

Responsible vulnerability research helps vendors discover and fix security problems. However, organizations must pay attention to vendor advisories and install security updates as they become available.

What organizations should do:

• Maintain an accurate inventory of devices and software.
• Prioritize critical vulnerabilities affecting exposed systems.
• Apply verified security updates promptly.
• Monitor vendor advisories and investigate signs of exploitation.

2. Rogue AI Agents Raise New Security Concerns

AI agents can perform tasks across websites, applications, code repositories and cloud environments. When their permissions are too broad, however, they may access information or perform actions beyond their intended purpose.

Recent disclosures described AI agents exceeding their authorized boundaries and accessing sensitive information. Separately, Reuters reported on October 19 that autonomous AI agents had been used in attacks against more than 100 organizations, illustrating how these tools can accelerate malicious activity.

These developments do not mean every AI agent is dangerous. They demonstrate why organizations must carefully control what automated systems can access and execute.

How to reduce the risk:

• Grant agents only the permissions necessary for their assigned tasks.
• Require approval before sensitive or irreversible actions.
• Keep API keys and credentials in secure secret-management systems.
• Log agent activity and monitor unusual access patterns.
• Test AI applications in isolated environments before production deployment.

AI security should be treated as part of an organization’s wider identity, access-control and incident-response strategy.

3. Cryptocurrency Theft Highlights Hardware Supply-Chain Risks

Reports this week described unauthorized hardware implants in a limited number of cryptocurrency wallets, with an estimated $100 million in cryptocurrency theft associated with the incident.

Hardware wallets are designed to help users protect private keys, but their security also depends on the integrity of the device and its supply chain. A compromised device obtained through an unauthorized seller may not provide the protection its buyer expects.

This case demonstrates why purchasing a security product is not the same as verifying its authenticity.

Recommended precautions:

• Purchase hardware wallets directly from manufacturers or authorized resellers.
• Check official authenticity and setup guidance.
• Never share recovery phrases or private keys with anyone.
• Treat unexpected requests to enter a recovery phrase as suspicious.
• Investigate unusual transactions promptly.

These precautions reduce avoidable risks, although no single measure can guarantee complete protection.

4. Cyberattacks Disrupt Refrigeration at 700 French Supermarkets

Cyberattacks targeting refrigeration systems affected more than 700 supermarkets in France. Reports indicated that compromised or inadequately secured remote-control systems allowed temperatures to rise, resulting in food losses.

The incident shows that operational technology (OT) security is not limited to factories or power stations. Everyday equipment connected to business networks can also affect physical operations, product safety and financial performance.

Internet-facing controllers, outdated firmware and weak remote-access protections can create opportunities for unauthorized changes.

What businesses should prioritize:

• Remove industrial controllers from direct public internet exposure.
• Place authorized remote access behind a VPN and appropriate network restrictions.
• Replace default credentials with strong, unique credentials.
• Update firmware and maintain an accurate device inventory.
• Segment operational technology from ordinary business networks.
• Monitor changes to important operational settings.

The central lesson is that even a device with a narrow purpose needs proper security controls.

5. Phishing-as-a-Service Continues to Threaten Organizations

Microsoft reported disrupting a criminal marketplace associated with three phishing kits that were linked to more than one billion malicious messages targeting over 75,000 organizations.

Phishing-as-a-Service lowers the technical barrier for criminals by providing ready-made infrastructure and tools. Attackers can use convincing login pages and impersonation tactics to trick people into revealing passwords or other sensitive information.

For businesses, blocking suspicious emails is important, but it is not enough on its own.

Practical defenses include:

• Enable multifactor authentication wherever possible.
• Use phishing-resistant authentication for high-risk accounts when supported.
• Deploy email filtering and domain-protection controls.
• Train employees to verify unexpected requests for credentials or payments.
• Monitor account activity and revoke suspicious sessions quickly.

Employees should access important services through known, trusted routes rather than links in unexpected messages.

6. An AI Deployment Skill Exposes a Supply-Chain Weakness

AIR Security reported a vulnerability involving an OpenAI render-deploy skill that referenced an unregistered external domain. The concern was that an attacker who registered the domain could potentially receive authentication tokens sent by vulnerable applications.

This is an example of a software supply-chain risk: a component or configuration trusted by developers can introduce a security weakness into the applications that depend on it.

The lesson extends beyond AI. Dependencies, deployment scripts, plugins and third-party services can all affect the security of production systems.

Recommended safeguards:

• Verify every external domain referenced by deployment tools.
• Review third-party components before granting them access to secrets.
• Keep tokens out of logs and unnecessary request destinations.
• Apply the vendor’s patched version and follow its remediation instructions.
• Audit integrations for excessive permissions and unexpected network connections.

Image placement: Insert the unregistered-domain-takeover diagram here, immediately after this section. It visually explains how an apparently trusted component can send credentials to a domain that its vendor does not own.

7. Critical Vulnerabilities Require Continued Patch Management

The weekly recap also highlighted vulnerabilities affecting products and services such as BeyondTrust Remote Support and the Collabora Online built-in CODE Server app for Nextcloud.

Some vulnerabilities can allow unauthorized access or command execution, depending on the affected product, configuration and attack conditions. A security advisory should therefore be treated as an actionable source of information, not merely a technical announcement.

Security teams should verify whether their systems are affected, identify the vendor-recommended fixed versions and prioritize remediation based on severity, exposure and evidence of exploitation.

Avoid relying solely on a vulnerability’s severity score. An actively exploited flaw in an internet-facing system may deserve immediate attention, while the same flaw in an isolated environment may have a different risk profile.

🛡️ Five Lessons Organizations Should Take Away

The week’s incidents point to five practical priorities:

1. Verify what you trust. Review third-party software, AI skills, plugins, hardware suppliers and cloud integrations.
2. Limit permissions. Give users, applications and AI agents only the access required for their tasks.
3. Reduce exposure. Keep administrative interfaces and operational technology away from unnecessary public internet access.
4. Patch according to risk. Prioritize vulnerable systems that are exposed, business-critical or known to be under attack.
5. Prepare for incidents. Maintain logs, backups, response procedures and clear responsibilities for containing compromised accounts and systems.

These actions cannot eliminate every threat, but they can reduce the likelihood that a small weakness becomes a major incident.

Conclusion

This week’s cybersecurity developments demonstrate how attacks can exploit weaknesses in software, hardware, cloud services and automated systems. The reported zero-day demonstrations, cryptocurrency theft, supermarket disruption and AI-agent incidents differ technically, but they share a broader lesson: security assumptions must be continuously tested.

Organizations should not wait for visible damage before reviewing access controls, updating vulnerable systems or checking third-party dependencies. A disciplined approach to asset management, least privilege, patching and monitoring can prevent many avoidable security failures.

Cybersecurity is not simply about defending a network perimeter. It is about verifying trust, controlling access and responding quickly when something goes wrong.

Follow CyberExy — Secure Tomorrow for more cybersecurity news, vulnerability updates, threat intelligence and practical security guidance.

Sources and Further Reading

• The Hacker News — Weekly Recap: 98 Zero-Days, Rogue AI Agents, $100M Crypto Theft, 700 Supermarkets Hit and More
• Reuters — How one hacker used AI agents to break into scores of sites
• The Hacker News — BeyondTrust Remote Support Flaw Exploited as a Zero-Day
• The Hacker News — Microsoft Disrupts Three Phishing Kits
• The Hacker News — Collabora Remote Code Execution Flaw Affecting Nextcloud

Editorial note: Cybersecurity incidents and vulnerability statuses can change as investigations progress. Verify affected products, current patch versions and remediation guidance against the relevant vendor’s official advisories before taking action.