Companies expand AI cyber defenses amid escalating attacks | Magica
Companies expand AI cyber defenses as attacks strain security teams
Editorial Team
••📖9 min read
Companies are expanding AI monitoring and patching as cyber threats strain security teams. Binance’s preliminary breach findings and attacks on aging refrigeration systems show why faster tools also need tighter access controls and investment in basic infrastructure.
Companies are moving AI into continuous monitoring and patching, while others are still building or testing defenses.
Binance says an internal AI tool helped attackers move between engineers’ computers before its wallet breach.
Rising budget expectations coexist with limited willingness to authorize fully autonomous defenses.
Companies are moving AI into around-the-clock monitoring and software patching as a surge in disclosed vulnerabilities strains security teams. Yet the same response must address compromised internal tools, aging equipment and permissions that can outlast an emergency shutdown.
German web hosting company Ionos, which serves seven million customers, says it changed its security practices after a July spike in vulnerabilities. It now uses AI-based monitoring and AI-assisted patching around the clock, according to a November 5 report. Travel platform Booking.com says hundreds of its engineers are building AI defense systems. Industrial group Siemens is testing advanced models for defensive security, while the Port of Rotterdam is working on an AI-powered cyber shield.
The business case is faster detection and repair, but these companies are at different stages: building a system, testing a model and operating continuous patching do not establish the same readiness or demonstrate the same results. None of these examples supplies a comparable deployment cost or measured return.
PwC’s company-reported survey findings show cyber-budget increase expectations rising from 78% to 84% among security and finance leaders. Source: PwC.
The cost of responding too slowly
The pressure reaches well beyond stolen data. Grocer Ahold Delhaize, operator of the Stop & Shop and Giant Food chains in the United States, suffered more than €200 million in sales losses after a mid-October intrusion and issued a profit warning, according to the November 5 report. It took websites and apps offline for a week. Some of its Belgian and Luxembourg stores were subsequently affected by refrigerator attacks this week.
Microsoft, Amazon and Alphabet rallied as investors bet Democratic election gains would restrain AI infrastructure spending. Strong cloud demand, existing commitments and risks to indebted operators complicate the hoped-for cash-flow relief.
Editorial Team
The report cites nearly 37,000 software or computer-system vulnerabilities recorded in the Common Vulnerabilities and Exposures registry in the third quarter of 2026, three times the year-earlier total. Disclosed flaws are a measure of remediation workload, not a count of successful attacks or proof that AI caused them.
There is also evidence of AI-assisted discovery creating an urgent defensive workload. The report describes the October 21 disclosure of two AI-discovered flaws in PHP, a programming language underpinning many websites. Internet infrastructure company Cloudflare said attempts to exploit the flaws increased from thousands to millions within hours of disclosure. Governments and companies took websites offline as a precaution; those attempted exploits do not measure successful breaches. Building-materials maker Holcim took its website offline that evening to install PHP fixes and restored it before midnight, showing that precautionary downtime could also be brief.
Cybersecurity company Claroty offers a narrower demonstration of AI’s effect on the time available to respond. After publishing research on refrigeration-controller weaknesses in August, its researchers used widely available AI tools to reproduce the flaws and control a test device following the October 11 attacks in France. The company said that reproduction cost $6 and took 28 minutes. That is a test-device result, not evidence identifying the tools used by the actual attackers.
Indeed, separate reporting on the refrigerator incidents describes familiar weaknesses: outdated remote-control modules and discontinued equipment. Danfoss said most affected systems were older products no longer receiving software or security updates; Copeland identified a vulnerability in discontinued controllers. Experts said there was no evidence AI powered the attacks. British cybersecurity officials described known vulnerabilities, direct internet exposure and default passwords in the French incidents.
Connectivity has an economic purpose: retailers monitor temperatures across stores, get spoilage alerts, reduce energy costs and arrange remote repairs. Securing or replacing exposed controllers is therefore part of the investment decision alongside buying AI defenses. Faster analysis alone cannot bring discontinued equipment back under security support.
Binance’s company-reported chronology traces the attack through engineering devices and an internal AI tool to wallet-signing operations. Findings remain preliminary. Source: Binance.
Binance’s AI tool was also part of the attack path
Binance, a cryptocurrency exchange whose internal wallet-signing service authorizes asset transfers, illustrates why buying better detection is only part of the response. In its preliminary investigation update, the company put the assets involved in its October 30 incident at approximately $6.34 billion. It said approximately $2.85 billion had been frozen or intercepted, with the amount still being verified and subject to market-price changes.
Binance said it would bear the full cost rather than pass losses to users, had set aside more than enough funds to cover them, and had replenished 98% of affected token assets by October 31. Replenishing customers and intercepting stolen assets are distinct measures; neither establishes a final net loss.
This is not its first wallet-security failure. In a May 2019 disclosure, Binance said attackers withdrew 7,000 bitcoin from a hot wallet after obtaining user credentials and passing its security checks. It promised to cover that incident through its SAFU protection fund. The two disclosures describe different attack paths.
Its latest account traces a sequence of failures. In June, an engineer seeking to install an AI coding tool downloaded a tampered software installer, exposing wallet-team information despite the company detecting the anomaly, replacing the device and rotating credentials. A fake job interview in September then induced another engineer to install malicious code on a work computer. Binance says endpoint security failed to detect that malware. On October 20, attackers used that compromised computer and an internal AI tool to take control of two colleagues’ devices, Binance said. A stolen token with permission to deploy software containers subsequently helped them reach wallet-signing operations.
Containment also depended on permissions already granted outside normal withdrawal systems. Binance said attackers abused the Ethereum blockchain’s EIP-7702 mechanism, which lets an account delegate execution permissions to a smart contract. Suspending withdrawals did not invalidate authorizations already recorded on the blockchain; separate transactions were needed to revoke them. The affected hot wallets used multi-party computation, meaning there was no single private key available for immediate revocation. Binance says it had to design a special revocation procedure, extending containment time.
Binance says it has rebuilt affected wallet systems and its internal AI tool, strengthened access controls, replaced devices and revoked affected credentials. It also plans more AI-driven detection spending and an independent security review. Its findings remain preliminary.
Binance says the FBI attributed the incident to North Korea and says its security team had prevented 57 suspected North Korea-linked incidents since the start of 2026. Those are claims in Binance’s account; the internal-tool compromise does not itself establish that an autonomous AI system conducted the attack. The company also proposes a security research fund with AI and security organizations, without specifying its size.
Earlier research suggests access control deserves particular scrutiny. An IBM-sponsored and analyzed study conducted by Ponemon Institute examined breaches at 600 organizations globally from March 2024 through February 2025. Thirteen percent reported breaches of AI models or applications; among that subgroup, 97% reported lacking AI access controls. This was a population of already-breached organizations, not an estimate of how frequently all companies suffer AI breaches, and it does not establish what caused Binance’s incident.
The same release supplies a counterweight to the risk findings: IBM reports that organizations using AI and automation extensively in security operations saved an average $1.9 million in breach costs and shortened the breach lifecycle by 80 days on average. The release does not specify that comparison group. These observational findings do not isolate AI’s causal effect or establish a return after deployment costs. IBM, a provider of AI, hybrid-cloud and consulting services, also has a commercial interest in adoption.
A growing market with an autonomy constraint
Security suppliers have a commercial stake in the response. The November 5 report cites McKinsey’s forecast that the global market for externally supplied cybersecurity products and services will rise from $223 billion in 2025 to $346 billion in 2029. That is a market projection, not evidence of returns from AI defenses.
PwC, the professional-services network that sells managed cybersecurity services, announced a cyber-defense collaboration with AI developer OpenAI in July. It described exploring the integration of advanced AI into managed services and controlled security workflows, including alert triage, investigations, vulnerability prioritization and remediation planning.
The relationship predates the cyber announcement. In May 2024, PwC said its US and UK firms had agreed to become OpenAI’s first reseller of ChatGPT Enterprise, its business AI product, and its largest user at that time. The cyber initiative therefore extends an existing commercial relationship.
The July announcement claims potential cost and speed advantages but supplies no prices, investment amount or measured customer comparison. Its emphasis on controlling inputs, outputs and user access also makes clear that this is a proposition about governed workflows, not simply handing security operations to a model.
PwC’s 2027 Global Digital Trust Insights survey, released October 1, captured responses from 3,934 business and technology executives across 71 countries and territories. Among surveyed security and finance leaders, 84% expected cyber budgets to increase over the next 12 months, compared with 78% in the previous survey—a rise of six percentage points. These are spending expectations, not completed purchases or a measure of how much budgets will grow.
Security leaders’ priorities and their willingness to delegate diverge: 58% ranked AI among their top cyber-budget priorities, but only 22% would authorize fully autonomous execution by AI defense agents. Reliability and technological maturity were a top-three barrier to greater autonomy for 55%; accountability and explainability were cited by 46%. Half of security leaders ranked attacks on AI systems among the threats they were least prepared to address. Only 39% of security, risk and operations leaders reported a fully formalized, integrated continuity plan addressing cyber risks.
People and infrastructure constrain deployment too. Forty-four percent of surveyed chief information security officers cited workforce skills in AI oversight and governance among their leading barriers to greater agent autonomy. Security and risk leaders also reported adopting multi-cloud or hybrid-cloud strategies (54%) and strengthening regional data and technology redundancy (47%) to reduce concentration risk. These are reported organizational responses, not evidence that any particular architecture prevents breaches.
PwC’s company-reported findings compare separate questions about AI spending priorities and autonomous execution; the percentages are not parts of a whole. Source: PwC.
The next test is recovery and control
Binance’s promised independent review is a concrete next checkpoint: it is intended to verify the findings and guide remediation. The consequential questions include how the internal AI tool allowed movement between devices and whether emergency procedures can revoke every relevant authorization quickly.
For other buyers, the decision is how to divide spending among AI assistance, human oversight, access controls and the replacement of unsupported infrastructure. The evidence needed is deployment-level measurement of containment and restoration times, total costs and the permissions granted to automated systems. Budget intentions and vendor promises cannot settle that allocation.
A symbolic strike by 11 Micron union officials left production running but exposed a larger dispute over who sets the rules for sharing the memory chipmaker’s AI gains.
Five technology groups have accumulated an estimated $3.2 trillion in future purchase and lease commitments. Filings show the funding pressure, but broad spending categories, long payment schedules and a methodology change limit what the total says about AI debt and growth.
PwC finds growth in a subset of AI-exposed junior job postings demanding senior skills, alongside pressure on entry-level opportunities. Workplace research suggests AI can support learning, but leaves employers to demonstrate how recruits acquire independent expertise.
CoreWeave bonds fell after the election of candidates favoring data center restrictions. Texas governor-elect Gina Hinojosa wants to extend an existing approvals pause until lawmakers pass protections, adding delivery uncertainty for an industry whose expansion depends on power and financing.
Walmart plans new permissions and a gateway for outside shopping agents, while Amazon blocks unauthorized tools and sells ads through its own assistant. The stakes are product discovery, customer data and whether delegated purchases add sales or displace advertising opportunities.
Apple reportedly cut some engineers’ daily AI allowances from $300 to $200 in October and is weighing deeper restrictions. The limits tighten control over model use, but unused allowances make the savings uncertain.
Factory is acquiring coding-agent collaboration platform HumanLayer after raising $200 million in September. It plans to integrate shared tasks and planning tools, while HumanLayer promises open source, refunds and at least 30 days of continued service.
A minister identifies Google-commissioned research behind the action plan’s £400 billion AI claim. The government’s central estimate is £120 billion, but different methods and adoption assumptions prevent a straightforward forecast-cut comparison.
Research has separated language models’ detection of altered internal activity from their ability to identify its content. Mechanistic studies and skeptical controls leave the interpretation disputed—and one intervention that improved detection made other measures of self-knowledge worse.
The IEA maps 134 policies governing data-center integration, arguing that credible project queues, coordinated tariffs and flexible connections can protect consumers while accommodating AI growth. Many measures remain too new to judge.
Three watchdog groups seek to halt Project Meridian, alleging that its contractor-heavy membership and closed meetings violate federal advisory committee law. MITRE’s invitation for outside ideas complicates the dispute over public participation.
Meta is seeking up to €9 billion in its first euro bond sale, drawing record orders while offering wider spreads than Alphabet. Its advertising business and large liquid reserves cushion an AI investment push that has overtaken quarterly operating cash flow.
Pangram is expanding its text-detection business into images and video. Browser checks and developer access broaden its reach, while company benchmark scores leave real-world error rates and costs unresolved.
About two dozen restrictive data center measures appeared to pass in initial returns. Texas governor-elect Gina Hinojosa’s promised construction pause adds a state-level constraint, though the capacity ultimately delayed remains uncertain.
Riot Platforms says its 15-year Corsicana lease with Anthropic will generate about $28 billion in revenue. The tenant owns and operates the compute, while Riot develops power infrastructure and funds module reimbursements through debt or credits against future rent.
Sonos plans limited U.S. early access for its conversational assistant, with a reported November 6 start. Users can return to older voice controls—a choice that also separates local audio processing from the new assistant’s cloud processing.
Tesla, SpaceX and AI infrastructure shares fell after Democratic election wins, while broader stocks rallied. Texas governor-elect Gina Hinojosa’s proposed data center pause puts a major construction pipeline in focus, but Republicans retain control of the Texas Legislature and federal agencies.
Capital One Ventures leads Sentra’s $150 million Series C and gains a board seat as the company expands AI-agent data controls. Sentra reports roughly 100 major enterprise customers and revenue in the tens of millions, but exact revenue, valuation and comparative costs remain undisclosed.
Google agreed to support Anthropic’s lease obligations at TeraWulf’s Kentucky campus, strengthening its bid for project financing. The support starts after delivery milestones, while debt terms and a guaranteed construction price remain unsettled.
