ServiceNow Buys Armis For $7.75B As Crypto Heists And AI Risks Surge In 2025

Cybersecurity news rarely lands in neat, separate buckets. This week proves why.
Here’s everything you need to know.

ServiceNow is making its biggest bet yet by buying Armis for $7.75B; crypto thieves are closing out 2025 with a record-breaking $2.7B haul; and OpenAI is openly admitting that AI browsers may never fully “solve” prompt injection.

Three different headlines, one shared theme: as software gets more connected and more autonomous, security risk scales with it.
 

TL; DR

 

• ServiceNow agreed to acquire Armis for $7.75B, strengthening its enterprise cybersecurity stack as AI-driven workflows and connected devices expand across modern organizations.

• Crypto thefts hit a record $2.7B in 2025, led by the $1.4B Bybit breach, reinforcing how exchanges and DeFi platforms remain prime targets.

• North Korean-linked hackers dominated major crypto heists again, with blockchain firms estimating at least $2B stolen to fund state-backed programs.

• OpenAI warned prompt injection attacks may never be fully eliminated in AI browsers, positioning them as a long-term security challenge requiring layered, continuous defenses.

 

ServiceNow’s $7.75B Armis Acquisition Pushes Deeper Into Cybersecurity

ServiceNow (a cloud-based platform) agreed to acquire nine-year-old cybersecurity startup Armis for $7.75 billion in an all-cash deal, marking its largest acquisition to date.


The deal is expected to close in the second half of 2026, per reports, as ServiceNow expands security workflows alongside its broader AI and platform strategy.

Armis recently raised a $435 million pre-IPO round, valuing it at $6.1 billion, making the acquisition a notable valuation jump. Armis co-founder and CEO Yevgeny Dibrov previously said the company aimed to go public in late 2026 or 2027, calling an IPO his “personal dream.”

ServiceNow reports that Armis has reached $340 million in annual recurring revenue (ARR), with year-over-year growth exceeding 50%, adding immediate scale to ServiceNow’s cybersecurity portfolio.

The acquisition also follows a busy year for ServiceNow, including its $2.85B acquisition of Moveworks and its agreement to acquire cybersecurity startup Veza for $1B.

From ServiceNow’s perspective, the Armis acquisition is intended to strengthen the cyber posture of modern enterprises, where AI, devices, and workflows are increasingly intertwined.

Although enterprise security is only one side of the risk equation, on the financial frontier, attackers are moving faster, bolder, and more profitably than ever.
 

Crypto Theft Hit A New High In 2025 As $2.7B Goes Missing

Blockchain-monitoring firms estimate that cybercriminals stole $2.7 billion in crypto in 2025, setting a new record for crypto theft.

The biggest incident was the breach at Dubai-based exchange Bybit, where hackers stole around $1.4 billion in crypto, and investigators (including the FBI, per reporting) linked the heist to North Korean government hackers.

Chainalysis and TRM Labs both estimated the $2.7B total, and De.Fi’s REKT database also aligned with that figure, reflecting a broad consensus across monitoring groups.

North Korean-linked hackers were again described as the most successful crypto thieves in 2025, with reports attributing at least $2B to them across the year’s thefts.

Other notable hacks cited in reporting included Cetus ($223M), Balancer ($128M), and Phemex (over $73M), reinforcing that exchanges and DeFi remain prime targets. The trendline is moving in the wrong direction: 2024 saw $2.2B stolen and 2023 saw $2B, making 2025’s number a new ceiling.

Chainalysis has also published additional analysis suggesting 2025 theft totals could vary by methodology, while still highlighting the outsized impact of the Bybit breach.

As these digital systems become more autonomous, security risks extend beyond stolen assets. They are embedded in how AI itself interacts with the open web.
   

OpenAI Says Prompt Injection May Never Fully Go Away For AI Browsers

OpenAI says it’s hardening its Atlas AI browser against cyberattacks, but it openly acknowledges that prompt injection is a persistent risk for AI agents operating on the open web.

In its own words: “Prompt injection, much like scams and social engineering on the web, is unlikely to ever be fully ‘solved,’” and OpenAI also conceded that “agent mode” in ChatGPT Atlas “expands the security threat surface.”

OpenAI says it views prompt injection as a long-term AI security challenge and will need to continuously strengthen defenses by accelerating patch cycles and scaling large-scale testing.

A key differentiator in OpenAI’s approach is its “LLM-based automated attacker,” trained with reinforcement learning to probe for new ways to sneak malicious instructions into an AI agent’s workflow.

OpenAI claims this RL-trained attacker could steer an agent into “sophisticated, long-horizon harmful workflows” across tens or hundreds of steps, including discovering strategies not seen in human red teaming or external reports.


The U.K.’s National Cyber Security Centre has similarly warned that prompt injection may never be totally mitigated, advising teams to reduce risk and impact rather than assume it can be “stopped.”

Wiz principal security researcher Rami McCarthy summed up the core risk model as: “A useful way to reason about risk in AI systems is autonomy multiplied by access.”