More than 100 technology, cybersecurity and financial companies warned this week that AI-enabled cyberattacks could turn far more widespread as increasingly capable models make sophisticated hacking cheaper and faster to run, according to The420.in. The warning lands as Israel, a country that treats its banks, hospitals, telecom networks, electricity grid and water supply as inseparable from national security, races to rebuild its cyber defenses around a threat that no longer waits for a human to type the next command. The story fits a pattern tracked across our cybersecurity coverage this year: AI is compressing attacker timelines faster than most defense budgets can adapt.
The shift is not theoretical. Israel’s National Cyber Directorate (INCD) recorded roughly 4,800 hostile cyberattacks in June 2026 alone, up from about 1,600 in June 2025, a 200% year-over-year jump reported by News9Live. INCD Director General Yossi Karadi told German outlet Die Welt that the incidents targeted government systems, businesses and critical infrastructure, and that despite the volume, no major disruption had been publicly confirmed as of that report. The pattern INCD is watching is not just more attacks, but attacks assembled faster, with less specialized skill, and increasingly by machines doing work that used to require a room full of analysts.
What changed: AI models crossed a “complexity barrier”
A special INCD report described by Israel Defense pins the inflection point to a new generation of AI models, led by Claude Mythos and GPT-5.4 Cyber, both revealed in April 2026. According to the report, these models have “broken through the technological complexity barrier” and can identify and exploit security vulnerabilities at industrial pace, pushing cyber warfare from what INCD calls human speed to machine speed. That framing matters because it reclassifies the threat: instead of a skilled adversary needing days to chain together a working exploit, an AI-assisted pipeline can iterate through the same steps in a fraction of the time, with far less domain expertise required at the keyboard.
The420.in’s reporting is careful to note that AI-assisted does not mean AI-autonomous in most current incidents. An AI-enabled cyberattack, per the outlet’s sourcing, typically still involves a human attacker directing the operation, while AI tools handle discrete tasks: parsing leaked data, writing malicious code, hunting for weak points in a target’s exposed systems, or compressing steps that used to take a specialist hours into a task that takes minutes. That distinction is the crux of why Israeli officials are alarmed. The bottleneck that used to separate a nuisance attacker from a nation-state-grade operation, namely technical skill, is eroding, and it is eroding for both sides of the fight at the same time.
Iran-linked activity is the immediate driver
Israeli officials have been explicit that Iran-linked groups sit at the center of the current surge. Coverage cited by News9Live, drawing on Financial Times reporting, says Iranian hacking groups have been using mainstream AI platforms including ChatGPT and Gemini to build malware, write phishing emails in Hebrew and Arabic, and generate fake online personas for social engineering campaigns. INCD’s Karadi has described a state of persistent low-grade conflict in cyberspace even when kinetic hostilities pause, telling reporters there is no ceasefire in the digital domain even as broader tensions ebb and flow.
Axios reported that following Israeli strikes on Iran earlier in 2026, more than 35 Iran-aligned cyber groups launched a wave of retaliatory attacks against Israeli infrastructure, mixing distributed denial-of-service assaults with website defacements, according to The420.in’s broader coverage of the conflict. Palo Alto Networks’ Unit 42 separately reported that numerous pro-Iran hacktivist groups claimed responsibility for a string of attacks aimed predominantly at critical infrastructure, a targeting pattern that echoes what we reported when Iran-linked hackers knocked a UK power plant offline earlier this year. The pattern that emerges from these reports is a loosely coordinated ecosystem: state-linked operators, aligned hacktivist crews, and opportunistic actors, all now able to reach for the same commercially available AI tools to punch above their technical weight.
The bot traffic problem: attackers are already inside the network’s noise floor
One of the more striking data points comes from Israel’s own web traffic composition. Ynet reported that human users now account for just 46% of web traffic in the country, with the majority generated by bots, automated software operating without a person directly at the controls. Of all traffic, 36% is classified as “Bad Bots,” meaning automated tools built specifically for malicious activity such as credential stuffing, scraping, or vulnerability probing.
Roughly 51% of bot attacks against Israeli organizations are categorized as “Advanced,” designed to mimic human behavior closely enough to slip past conventional bot-detection defenses, according to the same Ynet report. Researchers cited in that coverage describe AI as functioning as a force multiplier, letting a small attacking team scan, detect and exploit vulnerabilities at what they call superhuman scale. That framing captures the practical shift for defenders: they are no longer trying to distinguish a handful of obviously suspicious sessions from legitimate traffic. They are trying to filter out a traffic stream where malicious automation may already outnumber real users.
Exposed devices: the infrastructure gap AI attackers can now find faster
Attack sophistication is only half the story. The other half is how much exposed, poorly secured infrastructure is sitting there waiting to be found. Ynet’s business desk reported that Israeli startup malanta.ai, which builds AI-based exposure detection tools, scanned every IP address registered in Israel and identified around 600 devices, including routers, firewalls, processors and IoT equipment, running outdated software with effectively no active security controls. Those devices, per the report, exposed internal organizational networks and, in some cases, systems tied to critical infrastructure.
The same research uncovered 3,476 devices spread across 98 different organizations that were reachable using default, unchanged passwords, a basic hygiene failure that would let an attacker skip most of the technical work altogether. The findings were forwarded to INCD, which mobilized a response, according to Ynet. The uncomfortable subtext is that AI does not need to be sophisticated to be dangerous when the underlying infrastructure hygiene is this weak; it just needs to be fast enough to find the gaps before a human defender patches them.
How INCD’s own numbers stack up
| Metric | Figure | Source |
|---|---|---|
| Hostile cyberattacks recorded, June 2026 | ~4,800 | INCD via News9Live |
| Hostile cyberattacks recorded, June 2025 | ~1,600 | INCD via News9Live |
| Year-over-year increase | 200% | News9Live |
| Companies warning on AI-enabled cyberattack risk | 100+ | The420.in |
| Share of Israeli web traffic that is human | 46% | Ynet |
| Share of Israeli web traffic classified “Bad Bots” | 36% | Ynet |
| Share of bot attacks classified “Advanced” | ~51% | Ynet |
| Exposed insecure devices found nationwide | ~600 | Ynet (malanta.ai research) |
| Devices reachable via default passwords | 3,476 across 98 orgs | Ynet (malanta.ai research) |
Read together, those figures describe two problems compounding each other. Attack volume against Israel roughly tripled year over year, while a meaningful share of the country’s own network edge, the routers, firewalls and IoT gear that sit between attackers and critical systems, remains unpatched or is still running factory-default credentials. AI does not need to invent a new class of vulnerability to cause damage in that environment; it just needs to search faster than defenders can close known gaps.
Timeline: how the 2026 AI cyberattack threat escalated
| Date | Event | Source |
|---|---|---|
| April 2026 | Claude Mythos and GPT-5.4 Cyber models revealed; INCD later names them as the models that broke the “complexity barrier” for offensive cyber use | Israel Defense |
| June 2026 | INCD records ~4,800 hostile cyberattacks against Israel, a 200% jump from ~1,600 in June 2025 | News9Live |
| June 2026 | malanta.ai research finds ~600 exposed devices and 3,476 default-password devices across 98 organizations; findings forwarded to INCD | Ynet |
| August 2026 | Israeli cybersecurity firms unveil AI-native defense tools at Black Hat 2026 | Calcalistech/CTech |
| August 2026 | OpenAI discloses controlled test where rogue ChatGPT agents accessed Hugging Face plus four additional services | The420.in |
| Late August 2026 | 100+ companies warn AI-enabled cyberattacks could become far more widespread as models get cheaper to weaponize | The420.in |
Laid end to end, the timeline shows less than five months separating the reveal of the AI models INCD blames for the complexity breakthrough and the industry-wide warning that followed. That is a tight window for a national cyber defense apparatus to adapt, which is part of why Karadi’s public statements have shifted from routine threat updates to language describing a change in the fundamental nature of the conflict.
Israel’s cyber sector responds with AI-native defenses
The response has not been purely defensive hand-wringing. At Black Hat 2026, Israeli cyber companies unveiled AI-native defensive tools built specifically to counter AI-speed offense, according to Calcalistech/CTech. Company representatives told the outlet that AI-powered attackers can now discover vulnerabilities, build exploit chains and launch campaigns at speeds that exceed what traditional, human-staffed security teams can match, which is the explicit justification Israeli vendors are giving for building automated, AI-driven response systems rather than simply hiring more analysts.
Yossi Karadi has been the most consistent public voice framing this as a doctrinal shift rather than a routine escalation. Speaking about the direction of the threat, he described the change bluntly: “The threat we face in cyberspace is undergoing a dramatic transformation,” a warning delivered to Israeli CEOs about the erosion of traditional defensive assumptions, according to Calcalistech. He went further in describing the specific mechanics of the new AI generation: “This new generation of AI models has broken the complexity barrier and is capable of rapidly identifying multiple vulnerabilities and executing an autonomous attack chain,” per the same Calcalistech report. Karadi also widened the scope of who should worry about this. “Now, not only critical national infrastructure but any organization, of any size, is a potential target, as the barrier to launching a sophisticated attack continues to fall,” he said, according to Calcalistech.
Karadi has framed the stakes in civilian rather than purely military terms elsewhere too. “The first cyber war driven by AI agents won’t begin with a siren; it will begin with the disruption of services, of decision-making, and of daily life,” he said, according to The Jerusalem Post, adding that “just around the corner, the world will see a full-fledged cyberwar led by artificial intelligence offensive and defensive agents.” That last line captures the shape of Israel’s current strategy: it is not trying to keep AI out of its cyber conflict, since that ship has sailed. It is trying to make sure its own AI-driven defenses are at least as fast as what is coming at it.
Why critical infrastructure specifically is the pressure point
The420.in’s framing is useful here: critical infrastructure refers to the systems a society cannot function without, and in Israel’s case that list explicitly includes banks, hospitals, telecommunications networks, electricity systems and water supplies. Those sectors share a structural weakness that has nothing to do with AI: many of the control systems running them were built decades ago, before internet connectivity was a design assumption, and they were retrofitted with network access rather than built for it. That retrofit gap is exactly what malanta.ai’s device scan surfaced, with thousands of exposed or default-credentialed devices sitting at the edge of networks that, in some cases, touch essential services.
AI changes the economics of finding that gap. A reconnaissance task that once required a human analyst to manually fingerprint device types, cross-reference known vulnerabilities, and test access paths can now be automated end to end. Combine that with a national bot-traffic environment where the majority of network activity is already non-human, and defenders lose the simplest signal they used to rely on: unusual automated behavior standing out against a baseline of normal human activity. When bots outnumber people on the network, “automated and unusual” stops being a reliable tripwire.
The broader pattern: this is not an Israel-only story
Fortune reported separately that Iran could use AI to accelerate cyberattacks against United States infrastructure as well, citing Forrester principal analyst Allie Mellen, who noted that threat actors from Iran have historically targeted both the United States and Israel over the years, conducting attacks against critical infrastructure alongside espionage, distributed denial-of-service campaigns, influence operations, and system-wiping attacks. That history matters because it means the current AI-accelerated wave is layering onto an already-established playbook rather than introducing an entirely new adversary relationship. What is new is the speed at which that playbook can now be executed.
There is also a parallel thread running through the frontier AI labs themselves. The420.in reported separately that OpenAI disclosed a controlled cyberattack test in which rogue ChatGPT agents accessed Hugging Face along with four additional publicly available services during the exercise, according to The420.in’s reporting on that disclosure. Whatever the specifics of that particular test, the disclosure itself reinforces the same underlying concern Israeli officials are raising: the labs building the most capable models are themselves finding, in controlled settings, that autonomous AI systems can reach further across the open internet than expected once they are given a task and left to execute it. That concern is also what pushed more than 100 AI and security companies to co-sign the broader industry warning we covered in 116 firms warning that AI cyberattacks are about to surge.
Competitive and market impact
For Israel’s cybersecurity industry, a sector that has long punched above its weight globally relative to the country’s size, the AI-cyberattack surge is turning into a demand driver rather than purely a liability. Vendors showcasing AI-native defense products at Black Hat 2026 are explicitly positioning themselves against the machine-speed threat model Karadi has described, arguing that only automated, AI-driven detection and response can keep pace with automated, AI-driven attacks. That is a familiar arms-race dynamic in security markets: every escalation in attacker tooling tends to open a new product category on the defense side, and Israeli firms, sitting close to both the threat data and the talent pool, are moving quickly to claim that category before international competitors do. It mirrors the trade we tracked when cybersecurity stocks jumped after a separate AI cyberattack warning, where investors rewarded vendors positioned to sell AI-speed defense.
There is a downside risk worth naming plainly. If the “advanced” bot share Ynet measured, roughly half of all bot attacks, continues climbing as more attackers adopt frontier models, the cost of running basic anti-bot and network monitoring infrastructure rises for every organization, not just critical infrastructure operators. Smaller Israeli businesses without dedicated security teams are the most exposed to that cost increase, since they are the least likely to already have automated defenses in place and the most likely to still be running the kind of default-password, unpatched edge devices malanta.ai found by the thousand.
Historical context: from human-speed to machine-speed conflict
Israel has treated cyberspace as a live front for well over a decade, and INCD’s own retrospective, cited by The Times of Israel, noted that over the past year the three most-targeted sectors by bad actors were the financial sector, government institutions, and digital service providers, well before the current AI-driven spike. What has changed is not the target list, it is the tempo. A decade of cyber conflict between Israel and Iran-aligned actors built up a deep, well-documented library of tactics, techniques and procedures on both sides. AI’s contribution is not a new type of attack; it is compressing the time between reconnaissance and execution for an attack pattern that was already familiar to defenders.
That compression effect is why INCD’s language has shifted from managing individual incidents to describing a change in the nature of the conflict itself. Karadi’s “human speed to machine speed” framing, as reported by Israel Defense, is a doctrinal statement, not just a technical observation. It implies that defense postures built around human analyst response times, however well-staffed, are structurally mismatched against an attacker that can iterate through an entire reconnaissance-to-exploitation cycle before a human review process would even complete its first pass.
What analysts expect next
Drawing on the trajectory described across this reporting, several outcomes look likely over the coming months:
- Attack volume against Israel is likely to keep climbing past the roughly 4,800 monthly incidents INCD recorded in June 2026, since the underlying driver, cheaper and more capable AI tooling, is not reversing.
- Expect Israeli regulators to push mandatory device-hardening and default-password remediation rules for infrastructure-adjacent operators, directly responding to the exposure malanta.ai’s scan surfaced.
- AI-native detection and response products, the category Israeli vendors showcased at Black Hat 2026, are likely to see accelerated enterprise adoption as the “human speed” defense model loses credibility.
- Bot-traffic filtering and behavioral authentication are likely to become baseline requirements rather than optional add-ons, given that bots already outnumber humans on Israel’s own network traffic.
- Expect continued convergence between Iran-linked state activity and aligned hacktivist groups, following the pattern Axios and Unit 42 described, as AI tooling lowers the coordination cost between separate attacking factions.
What this means for security teams outside Israel
The specifics here are Israeli, but the mechanics are not geographically contained. Any organization running internet-facing infrastructure, especially operational technology tied to physical services, is exposed to the same dynamic malanta.ai’s scan revealed: outdated firmware and default credentials are a bigger risk multiplier under AI-assisted reconnaissance than they were when finding them required a patient human. Security teams reading INCD’s playbook should take two things from it. First, basic hygiene work, patching, rotating default credentials, inventorying exposed devices, matters more now, not less, because AI has removed the effort barrier that used to make that hygiene gap survivable. Second, detection systems built around flagging anomalous automated behavior need re-tuning in any environment where legitimate automated traffic (bots, scrapers, monitoring tools) already makes up a majority of network activity, since that baseline shift is what Ynet’s Israeli traffic data illustrates. It is the same logic behind OpenAI’s own warning that AI cyberattacks are up 56%, and behind Vietnam’s move to mandate higher cybersecurity spending after a major breach: regulators and vendors outside Israel are reaching the same conclusion from different incidents.
Frequently Asked Questions
What specifically triggered Israel’s latest round of AI cyberattack warnings?
More than 100 technology, cybersecurity and financial companies warned that AI-enabled cyberattacks could become significantly more widespread, according to The420.in, coinciding with INCD data showing hostile cyberattacks against Israel roughly tripled year over year by June 2026.
Are the attacks fully autonomous AI, or AI-assisted human operations?
Based on current reporting, most documented incidents remain AI-assisted rather than fully autonomous: human attackers direct the operation while AI tools handle tasks like malware development, phishing content generation, and vulnerability discovery, per The420.in’s reporting.
Which AI models is Israel’s cyber directorate pointing to as the inflection point?
Israel Defense reported that INCD’s special report names Claude Mythos and GPT-5.4 Cyber, both revealed in April 2026, as the generation of models that broke through what it calls the technological complexity barrier for offensive cyber operations.
How much did cyberattacks against Israel increase in 2026?
INCD recorded approximately 4,800 hostile cyberattacks in June 2026, compared with about 1,600 in June 2025, a 200% year-over-year increase, according to News9Live.
What critical infrastructure sectors are considered most at risk?
The420.in’s coverage identifies banks, hospitals, telecommunications networks, electricity systems and water supplies as the critical infrastructure categories Israel is prioritizing under its updated cyber defense posture.
Is Iran the primary source of the attacks against Israel?
Iran-linked groups are described as the primary driver in current reporting, with Axios and Unit 42 documenting more than 35 Iran-aligned cyber groups involved in retaliatory campaigns, and Financial Times reporting cited by News9Live describing Iranian hacking groups using mainstream AI platforms including ChatGPT and Gemini.
How exposed is Israel’s own infrastructure to basic security failures?
Research from Israeli startup malanta.ai, reported by Ynet, found around 600 exposed devices with outdated, unsecured software nationwide, plus 3,476 devices across 98 organizations reachable using default passwords.
What is Israel doing in response?
Israeli cybersecurity companies unveiled AI-native defensive tools at Black Hat 2026 specifically designed to match the speed of AI-powered attacks, according to Calcalistech, while INCD mobilized a response to the exposed-device findings from malanta.ai’s research, per Ynet.