the change is finally coming

Technological concepts & security analysis (containers vs. isolation)

Containers as a “security boundary” (questioned)

The video argues that containers have traditionally been viewed as a security boundary because they confine code such that malicious activity should remain inside the container. However, a rising number of kernel CVEs has led people to question whether containers still provide meaningful isolation/containment.

How containers work under the hood (mechanism)

A container is described as running an image’s program in an environment created using the Linux clone syscall (similar to fork, but with more control).

The key feature is namespaces, which provide separate views for different system resources, including:

• File system namespace (appears like a separate root via chroot-like behavior)
• User namespace (process appears as “root” inside, with fake UID mappings)
• PID namespace, network namespace, and others

Despite the “new computer/OS” illusion, the critical point emphasized is that all containers share the same underlying kernel.

Kernel CVEs and container escape risk

The video explains that if a vulnerability enables local privilege escalation (LPE), malware inside a container can exploit it and escape isolation, ultimately executing on the native host kernel.

This is framed as violating the container security model, because the isolation assumes the kernel remains uncompromised.

Why LPE exploitation is becoming more feasible

The video notes that LPEs are not new (it cites Dirty Cow as an early major example). The claimed shift is that AI accelerating vulnerability discovery increases the frequency of kernel vulnerabilities and introduces new classes.

• The video’s assertion: vulnerabilities may be emerging on the order of “every couple of months” rather than every one or two years.

It names two “groundbreaking” vulnerabilities/classes discussed in the video:

• copy-fail
  • Enables an arbitrary 4-byte write into the page cache
  • Uses socket + crypto-splice behavior
  • Can overwrite a setuid binary to gain root
• dirty-frag
  • Described as overwriting /etc/passwd
  • Followed by escalation to root

Additional vulnerability class names are also mentioned (without a deep dive), such as Fragnesia and Dirty Decrypt, suggesting ongoing emergence of exploit primitives related to sockets/buffers.

---

Main suggested mitigation: micro-VMs

Why kernel-updating alone may not be sufficient

Updating kernels helps, but with frequent CVEs there’s a race condition:

• Can defenders patch before an attacker uses an unpatched bug?
• Are patches available and deployed fast enough?

Micro VMs as stronger isolation

The proposed alternative is micro-VMs.

Key distinction:

• Containers: rely on kernel namespaces (shared kernel)
• Micro-VMs: use a VMM (virtual machine monitor) and hardware virtualization for hardware-level separation

If a process is compromised inside a micro-VM, it should not be able to affect other workloads/the host unless the attacker finds a vulnerability in the hypervisor/VMM.

---

Mentioned related security research angle (KVM)

The video cites a Google KVM CTF competition where escaping a provided VM earns large bounties.

It claims participants (example mentioned: Hayun Woo Kim) discovered multiple KVM vulnerabilities, emphasizing that hypervisors can also be vulnerable—while implying the attack surface is smaller/different than the full Linux kernel.

---

Sponsorship / product features mentioned (code review support)

Code Rabbit (sponsored)

The video includes a segment explaining how Code Rabbit helps with security-related development hygiene by improving PR review clarity.

Highlighted features:

• Automated code reviews for PRs
• A walk-through of what changes do, including diagrams/visual diagrams to show how subsystems interact and where trust boundaries lie
• A “Change Stack” view that breaks down PRs by layers/subsystems rather than only file diffs, helping reviewers see exactly where new code integrates (e.g., front-end to back-end routing responsibilities)

---

Source articles / influences

• The video says it was inspired by an article by Zengpeng Liao (host/source mentioned as “Depth First”), arguing that containers are no longer a security boundary in practice.

---

Main speakers / sources

• Primary speaker: The channel host / security practitioner (referred to as “my channel,” also discussing “Hackerman Sticks,” and providing explanations throughout)
• Referenced sources:
  • Zengpeng Liao — article at Depth First
  • Code Rabbit — sponsor (product features described)
  • KVM CTF / Google — referenced competition
  • Hayun Woo Kim — example researcher mentioned as having found multiple KVM vulnerabilities