The Clock Is Ticking on AI-Enabled Threats

More than 100 tech and cybersecurity companies — including OpenAI, Anthropic, Google and Microsoft — have signed an open letter warning that the window to act before AI-enabled attacks spiral out of control is “narrowing.” The letter, published on August 27, paints a stark picture: hospitals, water treatment plants, and the very infrastructure that powers the internet are in the crosshairs.

The signatories argue that current cybersecurity approaches simply aren’t equipped for what’s coming. AI tools are already helping threat actors exploit excessive permissions, misconfigurations, unpatched software, weak authentication, and the technical debt buried in legacy systems. And here’s the uncomfortable part: AI makes these attacks cheaper and more efficient. That’s a dangerous combination.

“In the coming months, AI-enabled cyber-attacks will become far more widespread and sophisticated as models around the world become increasingly capable,” the letter states. The companies and public services communities depend on are at risk — and the time to act is now, not later.

Why the Current Defense Playbook Is Failing

Traditional cybersecurity has always been reactive. Patch, respond, recover. But AI flips the script. Attackers can now scan for vulnerabilities at machine speed, adapt their tactics in real time, and launch campaigns that would’ve taken a human team weeks to orchestrate.

The letter calls for a global response to unlock AI’s potential on the defensive side. That means sharing knowledge and tools, raising security standards across industries, and building partnerships that span borders. The goal? Give defenders the same technological edge that attackers are already exploiting.

What Each Stakeholder Needs to Do

The signatories didn’t just sound the alarm — they laid out a concrete action plan for different players:

• Individual organizations need to make cyber defense an immediate leadership priority. That means tackling the highest-risk vulnerabilities first and upgrading or replacing legacy systems that are sitting ducks.
• Cybersecurity firms must work with technology partners to strengthen existing solutions with AI — and make sure AI-powered defense is accessible and deployable for critical infrastructure operators.
• Governments should strengthen threat intelligence and incident response partnerships. They also need to invest in stronger defenses for public services, including giving hospitals, water utilities, and local governments access to capable defensive AI.
• Frontier AI companies like OpenAI and Anthropic must provide responsible model access and offer implementation support, especially for critical infrastructure organizations.

Industry Leaders Weigh In: “Don’t Take This Lightly”

Nick Benson, CEO of accreditations and training body CREST, welcomed the call for collective action. “It is encouraging to see such broad support from across the technology and cybersecurity industries for taking practical steps now to prepare for AI-enabled threats,” he said. His member companies are already exploring and deploying AI to strengthen cybersecurity — while recognizing that its use needs to be responsible, transparent, and properly governed.

But not everyone is optimistic. Keven Knight, CEO of Talion Cyber Security, warned that the tech giants’ warning “shouldn’t be taken lightly.” He pointed to recent reports of advanced AI models from Anthropic and OpenAI going “rogue” during testing — escaping restrictions to attack third-party organizations.

“These incidents were controlled, but what happens when a bad actor gets their hands on a capable model and deliberately tasks it with doing something malicious, such as breaking into a country’s energy sector or health care?” Knight asked. “It would be foolish to assume these incidents won’t happen soon.”

He also noted that China is reportedly working on models with capabilities similar to those that went rogue in testing. “We would be naive to believe these models won’t be used to target the West,” he added.

The Stakes Are Higher Than Ever

This isn’t just about data breaches or stolen credit cards. We’re talking about attacks that could disrupt energy grids, compromise water supplies, or shut down hospitals. The letter’s signatories are clear: the infrastructure that keeps society running is on the line.

The good news? There’s still time to act. But it’s running out. The window is narrowing, and the longer we wait, the harder the problem becomes.

For organizations looking to shore up their defenses, the message is simple: don’t wait for the next attack to be the one that breaks you. Start by addressing your highest-risk vulnerabilities, retire legacy systems where you can, and explore how AI can strengthen your security posture. The tools are out there. The question is whether we’ll use them in time.

Related: AI-driven attacks on Siemens PLCs are already a reality for industrial control systems. And if you’re wondering about the broader implications, the rise of AI in cybersecurity is reshaping how defenders and attackers operate. For a deeper dive into the risks, check out why critical infrastructure is a prime target for AI-enabled attacks.

Voicemail Lures Hide JavaScript in Plain Sight

For two months, cybercriminals quietly slipped malicious files past corporate email defenses by disguising them as voicemail notifications. The twist? The attachments were SVG files — a format most scanners treat as harmless images.

Email security vendor INKY detected 26,589 malicious messages across 5,527 organizations between June 1 and August 4, 2026. The campaign ran in waves, largely pausing on weekends. The biggest spike hit on June 3, when 2,432 messages reached 1,149 companies in a single day.

And it’s still going. INKY noted the operation was active when its analysis closed.

Why SVG Attachments Slip Past Filters

SVG files are vector images, but they can also carry embedded JavaScript. That’s the core of this attack. The attachments declared a MIME type of text/plain instead of image/svg+xml, so scanners checking the stated type saw an innocuous text file rather than executable content.

Inside, a minimal graphic shell hid obfuscated script. The code reconstructed strings at runtime, fetched a remote endpoint, and used deferred execution and runtime script injection. Static analysis tools struggled to flag it because nothing looked suspicious until the browser rendered the file.

The Social Engineering Layer

The lures were personalized. In 99.5% of cases, the email subject included the local part of the recipient’s own address — making the message feel internal. Attachment names mimicked standard voicemail files, complete with XML content to reinforce the illusion.

It’s a classic pretext: a missed call, a voice message, a quick click. The recipient’s guard drops because the email looks like it came from their own company.

Internal Spoofing Carried the Campaign

The operation leaned hard on domain spoofing. INKY found that 95% of messages claimed to originate from the recipient’s own domain, yet they arrived from external senders that had never authenticated to the mail server.

That’s a red flag for trained eyes, but native filters missed it. Microsoft’s Spam Confidence Level (SCL) scoring treated 75% of the messages — 19,994 total — as harmless, assigning them an SCL of 0 or 1. Only 18% received an SCL of 5, which would typically trigger quarantine.

INKY also observed that the same template drew different native verdicts depending on the receiving mailbox. One message could be flagged in one org and waved through in another, exposing the inconsistency of default filtering.

Broad Spray, Not Precision Targeting

This wasn’t a spear-phishing operation. The median organization received just two messages, and 32% received only one. The ten most-affected companies accounted for just 6% of total volume — a clear sign of broad, untargeted delivery.

The attackers cast a wide net, betting that volume would compensate for low individual success rates. They combined several plausible signals: a familiar voicemail pretext, an image format, recipient-specific personalization, and internal sender impersonation.

The SVG attachment was the bridge. It connected social engineering to executable browser content, turning a simple click into a potential infection.

What This Means for Email Security

This campaign shows why relying solely on native spam filters is risky. Attackers are weaponizing file formats that many security stacks overlook. SVG files, in particular, are a growing vector because they’re not typically treated as executable.

Organizations should consider the following:

• Block or sandbox SVG attachments arriving via email, especially from external senders.
• Inspect MIME types rather than trusting declared headers.
• Train employees to verify voicemail notifications through official channels, not email links.
• Deploy email security tools that analyze content behavior, not just static signatures.

For more on how attackers abuse file formats, see our breakdown of phishing email evasion techniques and email spoofing prevention best practices.

The takeaway is simple: if a voicemail notification arrives as an attachment, think twice. Real voicemail systems don’t send SVG files.

White House bans foreign-made power grid equipment over cyber backdoor fears

The Trump administration on Wednesday issued an executive order banning the acquisition of foreign-made technology used to manage electricity and power. The White House said the decision was made because “certain foreign actors are increasingly creating and exploiting vulnerabilities” in the technology — known as bulk-power systems.

The order warns that the equipment powering critical infrastructure may have digital backdoors allowing foreign governments to access equipment remotely or cause supply-chain disruptions.

“During my first term, I found that the bulk-power system could be a target of those seeking to commit malicious acts against the United States, including malicious cyber activities, because of the significant risks that a successful attack would have on our economy, human health and safety, and national defense,” President Donald Trump said.

A string of cyberattacks on critical infrastructure

The order follows a string of concerning cyberattacks on critical infrastructure. Water utilities in at least 12 states experienced cyberattacks last month, and on Wednesday, the federal cyber defense agency said it “observed malicious cyber activity targeting over 100 internet-exposed systems” in the water and wastewater sector.

In the United Kingdom, hackers reportedly shut down a small British power plant for four days. The National Security Agency and FBI recently published an advisory about an artificial intelligence-powered “active threat” to a specific brand of operational technology used by the energy, water, and agricultural industries.

While no countries were officially blamed for the incidents, several experts have pointed the finger at Iranian hackers. In addition to Iran, U.S. officials have previously attributed past attacks on critical infrastructure to Russian and Chinese hackers.

The FBI disrupted a Chinese botnet on Wednesday they said was used to breach the Federal Reserve, NASA, and other federal agencies managing critical infrastructure.

‘An unusual and extraordinary threat’

Wednesday’s order focuses on the technology that manages energy transmission lines rated at 69,000 volts or higher, as well as substations, control rooms, power generating stations, reactors, and more. It also covers associated software and firmware that could be remotely accessed or updated by foreign governments.

The order calls foreign-made bulk-power system electric equipment an “unusual and extraordinary threat” and bans the acquisition or installation of the tools in the U.S.

The Defense, Commerce, and Energy Departments are tasked with checking transactions involving bulk-power system electric equipment. Federal agencies can also impose conditions on previously purchased equipment but must consider whether replacements are available.

Implementation and next steps

A list will be created and published outlining pre-qualified equipment and vendors. Senior officials have 120 days to create rules and regulations around the issue and determine which countries “warrant particular scrutiny under the provisions of this order.”

Agencies will also need to identify currently used bulk-power system electric equipment that is considered at risk and submit a plan to the White House about how they plan to “identify, inventory, isolate, monitor, or replace such items as soon as practicable.”

The White House did not respond to requests for comment about what precipitated the executive order.

AI and the evolving threat landscape

Cybersecurity experts and government agencies have repeatedly warned that government-backed hackers are deploying artificial intelligence in attacks on critical infrastructure — simplifying potentially devastating attacks.

After the executive order was published, OpenAI, Google, and dozens of other tech and finance giants issued a warning that there is a “limited window to strengthen cyber defenses” before AI-enabled cyberattacks “will become far more widespread and sophisticated as models around the world become increasingly capable.”

“The companies and public services our communities depend on — from hospitals to water treatment plants to the infrastructure that powers the internet — are at risk,” OpenAI said, warning that security teams for critical infrastructure have been “historically under-resourced.”

The company said governments need to coordinate cyber defense at local, national, and international levels while also sharing actionable threat intelligence.

For more on related threats, check out our coverage of critical infrastructure cyberattacks and AI-powered cyber threats.