Apple has issued a fresh round of threat notifications warning targeted iPhone users in 99 countries about mercenary spyware attacks. Security researcher Costin Raiu disclosed the wave on October 30, saying it was separate from August’s notifications and connected to new attacks.

The notification shared by Raiu states: “Today’s notification is being sent to targeted users in 99 countries, and to date we have notified users in over 150 countries in total.” The figures describe geographical reach, not the number of affected individuals or confirmed infections.

Apple’s notice further explains: “The extreme cost, sophistication, and worldwide nature of mercenary spyware attacks make them some of the most advanced digital threats in existence today.” It adds that Apple does not attribute the attacks or the notification to specific attackers or geographical regions.

Apple Warns iPhone Users in 99 Countries

Apple describes its threat notifications as high confidence warnings that an individual has been targeted by mercenary spyware. They are based solely on internal threat intelligence and investigations, although the company acknowledges that its assessments cannot achieve absolute certainty.

Receiving an alert does not establish that attackers successfully compromised the device. Reporting on the latest wave identifies journalists and documentary filmmakers among recipients, but Apple has not disclosed the total number of notified individuals or named the spyware involved.

The disclosures do not identify an exploit chain, vulnerability identifier, affected iOS versions, or a surveillance vendor. Consequently, attributing this wave to a named spyware product or claiming a specific infection mechanism would exceed the published evidence.

Apple withholds information about the signals triggering its alerts because revealing those details could help attackers adapt their techniques and evade future detection.

The company has issued these notifications multiple times annually since 2021, reflecting continuing investigations into highly targeted surveillance activity.

Apple delivers warnings through iPhone Lock Screen and Settings alerts, email notifications, and a banner displayed after users sign in to their Apple Account. Delivery methods vary depending on device model and software version.

Recipients should verify a warning by independently visiting account.apple.com rather than trusting unsolicited messages. Apple says genuine threat notifications never ask users to open files, install applications or configuration profiles, or disclose account passwords and verification codes through email or telephone calls.

Apple recommends enabling Lockdown Mode, which reduces the attack surface by restricting certain applications, websites, and device features. These restrictions can affect functionality, but provide additional protection against sophisticated targeting.

Notified users should also seek specialist assistance. Apple specifically directs recipients to Access Now’s Digital Security Helpline, which provides around-the-clock emergency support.

Expert assessment can help recipients evaluate the warning and decide on appropriate protective steps without assuming targeting necessarily resulted in a successful compromise.

Join 16,000+ SOC teams using ANY.RUN to streamline threat investigations and reduce manual effort. Explore for your team 

The post Apple Warns iPhone Users in 99 Countries of New Mercenary Spyware Attacks appeared first on Cyber Security News.

The group behind the Aur0ra ransomware has been using AI agents to move through victim networks, collect credentials and work its way up to domain administrator access, according to Sophos. AI-assisted intrusion workflow (Source: Sophos) The company, which tracks the group as GOLD OGRE, recovered artifacts from the Cursor AI coding assistant in late August that show the attackers handing the agents VPN logins, proxy access, targets and objectives. The agents then mapped the network … More →

The post Ransomware gang talked AI agents into doing much of its dirty work appeared first on Help Net Security.

Cisco Talos researchers have detailed VOLTA, a Windows information stealer that uses artificial intelligence to assess victims after collecting their passwords, account sessions, and cryptocurrency wallets.

The malware turns stolen information into ranked profiles, helping criminals identify people and assets worth pursuing. Self-named “VOLTA-AUTODOXER,” the malware surfaced through Talos’ CAIRN research effort.

Its AI component is optional, the underlying stealer can collect sensitive information even without an AI API key. Rather than directing the theft, the model performs the analytical work that an attacker would otherwise handle manually.

VOLTA AI Profiles Victims

VOLTA collects browser passwords, cookies, saved payment cards, cryptocurrency seed phrases, Microsoft OAuth tokens, SSH keys, cloud configuration files, and screenshots.

It also targets Discord, Telegram, Steam, Roblox, Slack, Outlook, Firefox, and Thunderbird data. An embedded cracking component attempts to unlock cryptocurrency wallet vaults during collection.

The malware extracts saved Wi-Fi passwords and then expands its findings using stolen account sessions.

It searches Gmail for shopping and payment receipts to establish home addresses, retrieves Uber Eats and DoorDash order histories, and gathers additional Discord profile details. Roblox inventories receive a live valuation through Rolimons.

These findings are assembled into Identity_Report.txt, including details such as employer, location, inferred birth year, and payment card information.

VOLTA sends the structured evidence to OpenRouter, instructing a language model to act as a forensic intelligence analyst preparing a target assessment.

The requested report includes “WHAT THEY’RE WORTH,” ranking assets by potential financial value, and “THEIR OPSEC,” assessing security practices and password hygiene.

The prompt also encourages connections between personal details and password patterns, extending the usefulness of the stolen records.

VOLTA defaults to anthropic/claude-opus-4.6. If responses contain refusal phrases such as “I can’t,” it can switch to moonshotai/kimi-k2, followed by google/gemini-2.5-flash. A temperature setting of 0.2 favors more consistent output, although it does not guarantee accuracy.

The returned assessment is saved as AI_Analysis.txt and packaged with the identity report in volta_dox.zip. Other collection stages can produce volta_log.zip and volta_doxnet.zip.

The AI request runs from the victim’s machine through their network connection before the results are sent out.

Talos identified six configuration slots populated during compilation, suggesting a builder service that creates customized binaries for customers.

This commercial model remains an assessment, not a confirmed sales arrangement. All observed network endpoints belong to third-party services rather than attacker-owned infrastructure.

Defenders should restrict which processes can launch Chrome with --headless, a behavior associated with VOLTA’s App-Bound Encryption bypass path.

Unexpected headless Chrome launches should trigger investigation. Hardware-backed credential storage is another recommended safeguard, talosintelligence said.

Because VOLTA deletes itself, investigators should preserve %APPDATA%\volta_debug.txt when available. This timestamped log can reveal what was collected and whether AI analysis succeeded.

Affected users should revoke compromised tokens and change passwords. If wallet seed phrases were exposed, funds should move to a new wallet.

Join 16,000+ SOC teams using ANY.RUN to streamline threat investigations and reduce manual effort. Explore for your team 

The post New AI-Powered VOLTA Malware Steals Passwords and Crypto Wallets to Identify High-Value Victims appeared first on Cyber Security News.

A single threat actor impersonated about 32 AI bots while scanning corporate websites for passwords, API keys, and cloud credentials.

The campaign used identities associated with ChatGPT, Claude, Perplexity, and other services to disguise malicious requests as trusted automated traffic.

On one day in October, the actor reached 48% of Cequence-protected organizations that recorded any AI bot traffic. Targets included businesses in financial services, retail, telecommunications, energy, and software.

The findings highlight a growing problem: attackers can copy an AI bot’s claimed identity without operating a legitimate AI service. A recognizable bot name alone does not establish trust.

Hackers Impersonate AI Bots

The scanner operated from a major cloud provider and rotated through identities linked to OpenAI, Anthropic, Meta, Perplexity, and others.

Rather than collecting ordinary web content, it requested roughly 90 file paths associated with exposed secrets. Targets included .env configuration files, cloud credential files, private SSH keys, and server environment information.

Each address sent between 50 and 190 requests in approximately eight seconds before moving on. The operation scanned targets sequentially, around the clock, and remained active from at least late September through late October.

Before checking sensitive paths, the scanner requested random, meaningless locations. This helped identify websites that returned responses even when requested files did not exist.

Several patterns exposed the impersonation. The scanner rotated evenly through all 32 identities, while the same machine claimed to represent different AI providers within seconds.

Its fake Claude identities also reused Anthropic’s crawler contact address, even when claiming to be another Claude agent. That inconsistency suggested copied identification strings.

Shared addresses and technical fingerprints connected activity across targets. According to Cequence, nearly all requests were blocked or failed, rather than producing confirmed credential theft.

The actor used more than 300 cloud addresses in one day and nearly 400 during one week. Addresses observed in late September did not reappear a month later.

Although threat feeds flagged many addresses, the scanner had usually switched infrastructure before those indicators became useful.

Its behavior, however, remained consistent across targets. Cequence describes behavioral analysis as a way to distinguish legitimate automation from abuse, rather than relying only on changing network identifiers.

The campaign also distorted AI traffic measurements. At several studied organizations, this impersonator generated more than 80% of traffic labeled as AI bot activity.

Separately, Axios reported that banks were restricting unvetted agents accessing customer accounts, often using credentials supplied by customers. That activity differs from this campaign’s credential hunting.

Cequence recommends verification through signed requests, published address ranges, or direct integrations. Defenders should also investigate assistants requesting configuration files, keep secrets outside web-accessible directories, and check whether staging or administration sites are publicly exposed.

The central lesson is that claimed identity and observed behavior must agree. Businesses should authorize verified agents for specific tasks, while monitoring their actions and rejecting requests that exceed those permissions, regardless of the familiar service name presented by the scanner.

Join 16,000+ SOC teams using ANY.RUN to streamline threat investigations and reduce manual effort. Explore for your team 

The post Hackers Disguise as 32 AI Bots to Scan Corporate Networks for Passwords and Cloud Credentials appeared first on Cyber Security News.

Scanning activity targeting BeyondTrust Remote Support appliances raises concerns that exploitation of CVE-2026-102764 is expanding beyond targeted intrusions to opportunistic attacks.

Previdian reported two probes against its decoy systems in the United States and Europe, while emphasizing that neither observation demonstrated successful exploitation.

The vulnerability previously enabled zero-day attacks against real appliances. However, the latest telemetry provides an early warning of broader interest rather than conclusive evidence that mass exploitation is already underway.

Hackers Scan BeyondTrust Remote Support Systems

The supplied monitoring screenshot records requests on November 1, 2026, at 09:06 and 10:54 UTC. Both originated from 189.24.79.18, which the monitoring platform geolocated to Hong Kong, and reached Remote Support decoys in Germany and the United States, respectively.

Both requests targeted POST /api/start_session. Previdian supplied the associated parameters as issue_menu=1&id=1&c2cjs=1&platform=thin.

These request details offer defenders a concrete pattern to examine when reviewing traffic directed at exposed Remote Support appliances.

The German decoy recorded a curl/8.5.0 user agent, while the American sensor logged a browser-style user agent identifying Windows and Chrome.

Those strings describe how the clients presented themselves; they do not independently establish the software or operating systems actually used.

Previdian said the traffic resembled a BeyondTrust Metasploit module currently under review, although it was not an exact match. That resemblance suggests possible public exploit experimentation, but does not prove which tool generated the requests or establish a connection to previously identified attackers.

The source address’s reported location likewise should not be interpreted as attribution. Most importantly, the sensors were decoys, and Previdian reported no successful exploitation from these attempts.

BeyondTrust disclosed CVE-2026-102764 on October 15. Its advisory describes insufficient validation of message origins within internal Remote Support components, allowing a remote attacker to perform unauthorized file operations on an affected appliance.

The flaw affects Remote Support 26.2.3 and earlier unless patched. Canada’s Cyber Center classifies it as authentication bypass by spoofing and warns that exploitation requires neither credentials nor user interaction, potentially enabling arbitrary file uploads and appliance compromise.

According to the supplied account of Mandiant’s findings, ARMORY CASTLE exploited the vulnerability in late August 2026. BeyondTrust identifies the group as a suspected China-linked espionage actor previously tracked as UNC5221 and confirms it exploited the vulnerability before disclosure.

CISA subsequently added the flaw to its Known Exploited Vulnerabilities catalog on October 16, according to its published alert. This distinguishes confirmed historical exploitation from the newer decoy observations, which indicate probing rather than verified compromise.

Organizations operating affected self-hosted appliances should prioritize the BT26-05-RS patch. BeyondTrust provides the patch for versions 24.3.1 through 26.2.3; older deployments require an upgrade before installing the patch. Remote Support SaaS customers were patched by October 9.

CISA also recommends restricting external access, limiting administrative interfaces to trusted networks, and investigating potentially compromised appliances. Relevant evidence includes unexpected privileged sessions, unfamiliar scripts, unauthorized account changes, and logging gaps.

Where compromise is suspected, organizations should preserve forensic evidence and isolate affected systems. Confirmed intrusions require recovery from a known-good state, patching, credential rotation, and revocation of active support sessions.

Join 16,000+ SOC teams using ANY.RUN to streamline threat investigations and reduce manual effort. Explore for your team 

The post Hackers Scan BeyondTrust Remote Support Systems for Actively Exploited Zero-Day Flaw appeared first on Cyber Security News.

Ivanti released security updates to patch two more Connect Secure (ICS) remote access gateway flaws that have been chained in zero-day attacks since at least October 1. [...]