# One Request Away: The Week's Exploits Prove the Attack Surface Is Already Inside Your Perimeter
The most unsettling thing about this week's vulnerability haul isn't any single bug. It's the ratio. RCE from a malformed request. Key theft from a fake prompt. Security tools disabled from a driver that shipped years ago. The attacker's investment keeps shrinking while the defender's blast radius keeps growing — and this week crystallized how badly that asymmetry has shifted.
---
## SonicWall's Edge Problem Is Everyone's Edge Problem
SonicWall disclosed zero-days affecting its SMA (Secure Mobile Access) appliances — devices sitting at the network perimeter, trusted to gatekeep remote access. If that sounds familiar, it should. This is not the first time SonicWall gear has been the entry point. In 2021, ransomware crews weaponized a heap-based buffer overflow in SMA 100 devices before patches were available. In 2023, a pre-authentication RCE landed in the same product line. The pattern is consistent enough that CISA has had SonicWall vulnerabilities on its Known Exploited Vulnerabilities catalog multiple times.
What makes perimeter appliance 0-days uniquely dangerous isn't just the exposure — it's the trust. These devices sit between the internet and everything else. Compromise one, and you're not an outsider anymore. You're on the network side of the firewall, authenticated by proximity, with a clean path toward domain controllers, file servers, and backup infrastructure. Ransomware operators understand this. So do the nation-state groups that frequently precede them.
Defenders running SonicWall — or any SSL-VPN appliance — should treat this as a recurring structural problem, not a one-time patch event. Segmentation of the network directly behind these devices is non-negotiable. If you're using the default "trusted" zone model where the VPN drops users into an open internal segment, you're one successful exploit away from a lateral movement chain that bypasses your EDR entirely.
---
## SharePoint: The Sprawl That Keeps Giving
A new SharePoint zero-day surfaced this week, adding another entry to what is quietly becoming one of Microsoft's more persistent attack surface problems. SharePoint runs in a staggering percentage of enterprise environments. It integrates with Active Directory. It's often over-permissioned. And because it's Microsoft, it's treated as inherently trustworthy infrastructure — which makes it exactly the kind of system adversaries want to weaponize.
The technical specifics matter less than the structural reality: SharePoint 0-days translate rapidly into domain-level access. Microsoft has fixed SharePoint RCEs before — CVE-2023-29357 was a clear example where a CVSS 9.8 flaw allowed privilege escalation without any credentials. The exploitation window between disclosure and mass weaponization on these Microsoft enterprise products has compressed dramatically over the past three years as exploit tooling has become commoditized.
Organizations relying on SharePoint should audit exposure now: what SharePoint instances are internet-facing? What permissions do service accounts running SharePoint carry? Are SharePoint servers segmented from domain controllers? These aren't new questions, but they require fresh answers every time a 0-day drops.
---
## WordPress RCE: The Platform That Never Stops Bleeding
WordPress remote code execution landed again. To regular HackWire readers, this may feel like a standing item — and it is. WordPress powers roughly 43% of every website on the internet, which means any RCE in its ecosystem is a mass casualty event in scope, even when attribution is murky and the CVE looks mundane.
The specific attack surface varies: sometimes it's a plugin (the WP File Manager RCE in 2020 was actively exploited within hours of disclosure). Sometimes it's a theme. Occasionally it's core. What never varies is the exploitation speed. WordPress targets get compromised fast because scanning tools are purpose-built for it, the user base includes enormous numbers of sites running unpatched plugins indefinitely, and the reward — web shell access to a public-facing server — immediately feeds SEO spam operations, credential phishing infrastructure, or cryptomining.
If you run WordPress at scale, the minimum hygiene bar is automated plugin updates, a WAF rule blocking known exploit signatures, and integrity monitoring on the wp-content directory. If you manage WordPress for clients and you're still manually patching: that's the gap adversaries are walking through.
---
## Fake Prompts and Real Key Theft: AI Services Under Fire
The AI attack surface generated its own entries this week, with reports of prompt injection and service-level attacks against AI-powered tooling. The mechanism is becoming familiar — an adversary crafts input that the AI processes as an instruction rather than data, then pivots that capability toward credential theft, context extraction, or manipulation of downstream actions the AI is authorized to perform.
This is not science fiction threat modeling anymore. Prompt injection has been demonstrated against real deployments: customer service bots redirected to exfiltrate conversation history, code assistants tricked into including malicious dependencies, AI-powered email tools manipulated into forwarding messages to attacker-controlled addresses.
The stolen keys angle matters separately. AI service APIs — OpenAI, Anthropic, Azure OpenAI — carry real monetary and access value. Leaked API keys get scraped from public GitHub repos constantly and monetized within hours, either through resale or direct abuse for compute-heavy operations. Any organization integrating AI services needs secret scanning in CI/CD, rotation policies for AI API credentials, and spend alerting to catch unauthorized usage before it becomes a large bill or a data exposure.
---
## Old Drivers Still Kill
Vulnerable drivers from years past continued to appear in active attack chains this week. Bring Your Own Vulnerable Driver (BYOVD) attacks have become a standard ransomware pre-deployment technique — attackers load a legitimate but exploitable signed driver to kill EDR processes before the ransomware payload runs. The defensive implication is brutal: your endpoint security product can be disabled by software the operating system trusts.
Microsoft's Vulnerable Driver Blocklist exists specifically for this, but it requires explicit enablement and isn't applied retroactively in all configurations. Checking whether it's active across your fleet should be immediate.
---
## HackWire Analysis
The through-line this week is the economics of access. Exploits are scaling down in complexity while scaling up in consequence — and the gap between "bug disclosed" and "bug weaponized" has effectively closed for high-value targets.
What's underreported in the standard vulnerability recap framing: the compounding effect. None of this week's bugs operates in isolation. A SonicWall edge compromise hands attackers an authenticated internal position. From there, a SharePoint RCE escalates privileges. A vulnerable driver clears the EDR. Ransomware deploys. Every link in that chain was available this week, simultaneously, from publicly disclosed vulnerabilities or in-the-wild 0-days.
The defenders who will fare worst are the ones treating each bug as a standalone patch ticket. The defenders who will fare best are the ones asking: if my perimeter appliance is compromised right now, how far can an attacker get before detection? If the honest answer is "very far," the patch queue is less urgent than the architecture.
The AI attack surface deserves particular attention because it's the newest dimension and the least understood operationally. Most organizations have connected AI tooling to real credentials, real data stores, and real action capabilities without building adversarial assumptions into their trust models. Prompt injection is not a theoretical future risk. It's this week's attack category.
The uncomfortable truth about weeks like this one: they're not exceptional. They're the baseline now.
— HackWire Editorial
---
## Related Coverage