We’re Tracking the Future of Security

Category: News

Current, event-driven reporting, announcements and industry developments.

  • Anthropic Says Seven China-Based AI Labs Ran Industrial-Scale Model Distillation Attacks Against Claude

    Anthropic Says Seven China-Based AI Labs Ran Industrial-Scale Model Distillation Attacks Against Claude

    Anthropic said it identified and disrupted industrial-scale “distillation” attacks against its Claude models originating from seven China-based AI labs, according to a threat intelligence report the company released on September 11, 2026, and covered by The Hacker News. The company named Alibaba, Moonshot AI, DeepSeek, Xiaomi and Zhipu, also known as Z.ai, among the labs involved, saying they routed user conversations through Claude and used its responses as training data to improve their own models in violation of Anthropic’s terms of service.

    Anthropic said the campaigns specifically targeted Claude’s most commercially valuable capabilities, including agentic reasoning, coding and tool use, and that some of the intercepted exchanges contained sensitive information belonging to individual users, multinational companies and state-affiliated actors. According to reporting on the report, Moonshot alone routed nearly 300,000 customer requests to Claude over a 10-day period through a network of more than 5,000 fraudulent accounts, with total distillation activity exceeding 23 million exchanges between May and July.

    Anthropic said labs used commercial proxy services and account networks to evade detection, allowing new fraudulent accounts to replace banned ones without interruption. The disclosure adds to a pattern of Western frontier labs publicly accusing competitors of extracting model capabilities without authorization, a dynamic that is intensifying as US and Chinese AI companies compete for enterprise and developer market share.

    Why it matters

    Routing third-party traffic through a commercial model’s API to harvest its outputs is itself a data-handling risk for the end users caught in the middle, since their prompts and outputs can transit infrastructure they never consented to; enterprises building on any AI API should ask vendors directly what abuse-detection and data-isolation controls apply to their traffic.

    Sources

  • Diesel Spill at New Jersey AI Data Center Draws Environmental Scrutiny

    Diesel Spill at New Jersey AI Data Center Draws Environmental Scrutiny

    An estimated 5,000 gallons of diesel fuel spilled from a storage tank at an Equinix data center in Secaucus, New Jersey, on the evening of Friday, September 11, 2026, and flowed into Anderson Creek, a tributary of the Hackensack River, according to CBS News New York, which cited the New Jersey Department of Environmental Protection. The facility, which is under construction and expansion, uses diesel tanks to power backup generators. Cleanup crews have used vacuum trucks to remove a mixture of fuel and water from the creek since the spill was detected.

    Equinix said in a statement that it detected the fuel release Friday, stopped it, and immediately contacted relevant agencies. “We are working closely with those agencies, local officials, and community stakeholders to ensure a thorough remediation, while also conducting a full investigation to determine the root cause and prevent recurrence,” the company said, according to CBS News New York. Bill Sheehan of the environmental group Hackensack Riverkeeper told CBS he was informed a computer glitch triggered the spill.

    The incident has renewed local debate over the rapid expansion of AI-driven data center capacity in the New Jersey Meadowlands, with advocacy groups using the episode to press for a statewide moratorium on new data center construction, citing the sector’s growing power and infrastructure footprint.

    Why it matters

    Backup-generator fuel systems are a routine part of data center resiliency design, but this incident is a reminder that the same infrastructure meant to protect uptime carries its own environmental and physical-security exposure that facility teams need to monitor, not just the IT floor.

    Sources

  • Musk Proposes Cross-Lab AI Model Testing, Including Chinese Developers

    Musk Proposes Cross-Lab AI Model Testing, Including Chinese Developers

    Elon Musk urged the world’s leading AI developers, including several Chinese labs, to submit their models to safety testing by competitors, speaking at the All-In Summit in Los Angeles on September 15, 2026, according to CNBC. Musk said xAI, OpenAI, Anthropic, Google, Meta and “three or four of the leading Chinese companies” should allow rivals to run a “test harness” on their models to flag safety concerns before release. “Instead of grading your own homework, you would at least have competitors grading your homework and raising the alarm if they see concerns,” Musk said, according to CNBC.

    The proposal came as debate intensified following a weekend in which Anthropic CEO Dario Amodei called for the industry to slow the pace of frontier model development, a position Musk and OpenAI’s Sam Altman both publicly backed. A spokesperson for China’s Foreign Ministry dismissed the broader push for a slowdown as “fear mongering,” according to remarks reported by the South China Morning Post.

    The exchange illustrates how quickly a safety debate largely confined to AI labs has spilled into geopolitics, with implementation of any cross-lab testing regime facing significant hurdles given the commercial rivalry between the companies involved and the strategic sensitivity of sharing frontier models across the US-China divide.

    Why it matters

    A cross-lab testing regime, even a voluntary one, would be a meaningful departure from today’s self-graded AI safety disclosures; whether it goes beyond a talking point will depend on whether any lab actually opens model access to a named competitor.

    Sources

  • Nvidia’s Huang Breaks With Anthropic, OpenAI Chiefs Over AI Safety Pacing at Dreamforce

    Nvidia’s Huang Breaks With Anthropic, OpenAI Chiefs Over AI Safety Pacing at Dreamforce

    Nvidia CEO Jensen Huang broke publicly with Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman on artificial intelligence safety policy during Salesforce’s Dreamforce conference in San Francisco on September 15, 2026, according to CNBC. Amodei opened the event alongside Salesforce CEO Marc Benioff, days after publishing an essay urging the industry to slow the pace of frontier model development. Huang, appearing later in the same keynote, said market forces were sufficient to ensure AI safety and rejected calls for a deliberate slowdown, according to CNBC.

    The clash followed a weekend in which Amodei proposed what he described as a framework for moderating capability gains without “sacrificing commercial advantage or the United States’ lead in AI,” CNBC reported. Altman, appearing separately with Benioff for an afternoon fireside chat, said safety and monitoring work must come before further capability gains, telling the audience “there should be no qualifier on that,” according to CNBC’s account of his remarks.

    Dreamforce, which draws roughly 50,000 attendees to downtown San Francisco annually, became an unusual venue for the dispute to play out in front of an enterprise software audience rather than specialist AI trade press.

    Why it matters

    For security and IT buyers evaluating AI-driven tools, a public split among frontier-model builders over safety pacing is a signal to keep independent evaluation and guardrails in-house rather than assume vendor consensus on responsible deployment.

    Sources

  • Axoflow Launches AxoDetect to Run Threat Detection Inside the Security Data Pipeline

    Axoflow Launches AxoDetect to Run Threat Detection Inside the Security Data Pipeline

    Axoflow has launched AxoDetect, a new capability that runs Sigma detection rules directly in its data pipeline rather than after logs land in a customer’s security information and event management (SIEM) platform. The feature, now in early access, sends triggered alerts on to the customer’s SIEM while routing full-fidelity logs to AxoLake, the company’s own storage layer, which Axoflow says lets organizations retain complete audit trails at lower cost than storing everything in a traditional SIEM.

    Axoflow, founded by Balazs Scheidler, creator of the open-source syslog-ng project, markets its platform as a security data layer that discovers, classifies, parses, normalizes and reduces security telemetry before it reaches downstream analytics tools. AxoDetect extends that existing pipeline to include in-stream detection logic ahead of the SIEM.

    The company is demonstrating AxoDetect this week at Splunk’s .conf26 conference in Denver, running September 14-17, at booth P1, and plans to show the capability again at the Gartner Security & Risk Management Summit in London later this month.

    Why it matters

    In-stream detection is a direct response to SIEM cost pressure: if triage logic can run before ingestion, organizations can route only actionable alerts to their most expensive analytics tier while still retaining full logs elsewhere for investigation.

    Sources

  • Group-IB Details ‘JWR’ Phishing Kit Powering Smishing Triad Bank-Fraud Campaigns

    Group-IB Details ‘JWR’ Phishing Kit Powering Smishing Triad Bank-Fraud Campaigns

    Group-IB has published a technical analysis of a phishing kit it calls JWR, used by an operator cluster it tracks as “Outsider” within the broader Smishing Triad ecosystem. In the campaign Group-IB analyzed, victims received fraudulent SMS messages impersonating official entities and were directed through a short link to a disposable domain, where a multi-stage form walked them through submitting identity details, payment card information, and one-time passcodes, sometimes for a second bank or a digital wallet.

    According to Group-IB, the kit’s landing page is built as a Vue 2 single-page application wired to a dedicated Web Worker that maintains a binary WebSocket channel, encrypted with an AES-256-CTR envelope, alongside a two-second HTTP long-poll fallback, letting a human operator monitor and steer victim sessions in near real time. Group-IB said code strings tie the kit to infrastructure known in operator circles as the trsb.top family.

    Group-IB, along with prior public reporting from Silent Push, Palo Alto Networks’ Unit 42, Resecurity and Fortra, has linked the wider Smishing Triad ecosystem to more than 194,000 malicious domains registered since 2024 spanning over 121 countries. Group-IB researcher Mohammad Gamal Younis authored the analysis, which included indicators of compromise for defenders.

    Why it matters

    Real-time, operator-steered phishing sessions raise the bar for defenders beyond static blocklists, since the kit adapts its lure mid-session to whatever the victim’s device or bank actually shows; financial institutions and their customers benefit most from the published indicators of compromise, not from generic awareness alone.

    Sources

  • Knightscope Debuts Full Autonomous Security Force Lineup at GSX 2026

    Knightscope Debuts Full Autonomous Security Force Lineup at GSX 2026

    Knightscope brought its full Autonomous Security Force lineup together under one roof for the first time this week at GSX 2026, the Global Security Exchange, at the Georgia World Congress Center in Atlanta. The Sunnyvale, California-based company’s booth displayed its K7 and K5 Autonomous Security Robots, the K1 Hemisphere, K1 Capsule and K1 Super Tower stationary units, a first public look at its H1 Augmented Security Agent prototype, and Signals, the company’s forthcoming command-and-control software platform.

    Alongside the live debut, Knightscope published “Human Capabilities,” the fifth in a series of field briefs describing its ASF-7 model, which pairs autonomous machines and software with licensed security agents employed directly by the company rather than through third-party staffing contractors. The company said its agents handle alert verification, executive protection, armed and unarmed security response, and de-escalation, according to the field brief.

    Knightscope said it currently serves 434 clients across 42 states, with its machines logging more than 4.4 million autonomous hours in the field. GSX 2026, which ran September 14-16, drew more than 14,000 security professionals, according to show organizers cited in the company’s announcement.

    Why it matters

    Knightscope’s move to combine hardware, software and directly-employed human responders under a single contract structure is a bet that integrated accountability, not just more autonomous hardware, is what will differentiate robotic security offerings going forward.

    Sources

  • Acronis Backup Plugins for cPanel and Plesk Carry Actively Exploited Privilege-Escalation Flaw

    Acronis Backup Plugins for cPanel and Plesk Carry Actively Exploited Privilege-Escalation Flaw

    Acronis has disclosed a high-severity local privilege-escalation vulnerability in its backup plugin for cPanel, WebHost Manager (WHM) and Plesk, warning that it has detected exploitation in the wild. The flaw, tracked as CVE-2026-87886, carries a CVSS severity score of 7.8 and allows a low-privileged attacker to increase their permission level on a vulnerable Linux server without user interaction, potentially enabling access to or modification of sensitive data, according to Acronis’s advisory.

    The plugins connect the widely used cPanel and Plesk hosting control panels to Acronis’s backup infrastructure, letting administrators back up and restore websites, databases, mailboxes and hosting accounts. Acronis said exploitation “has been detected in the wild in limited, targeted attacks against Acronis Backup plugin for cPanel & WHM deployments,” though it told BleepingComputer the assessment is based on a single report from a potentially affected customer.

    Affected versions include the Acronis Backup plugin for cPanel & WHM prior to build 1.9.3.1021, fixed in version 1.9.3 HF3, and the Acronis Backup extension for Plesk prior to build 1.8.11.638, fixed in version 1.8.11. Acronis is urging all users of the affected integrations to update immediately.

    Why it matters

    Backup plugins run with elevated privileges by design, which is exactly what makes a privilege-escalation flaw in one valuable to an attacker who already has limited server access; hosting operators should treat the patch as time-sensitive rather than routine.

    Sources

  • NATO Fighter Jets Shoot Down Drone That Crossed Into Lithuanian Airspace From Belarus

    NATO Fighter Jets Shoot Down Drone That Crossed Into Lithuanian Airspace From Belarus

    NATO fighter jets shot down a drone that entered Lithuanian airspace overnight, officials said Tuesday, in the latest of a series of unidentified drone incursions into Baltic airspace since Russia’s full-scale invasion of Ukraine. Lithuania’s National Crisis Management Centre logged the intercept at 01:19 local time near the village of Pratkunai in Kaisiadorys district, roughly 90 kilometers northwest of the capital Vilnius. No injuries or damage were reported.

    Lithuanian authorities said the drone most likely crossed into the country from neighboring Belarus. Italian Air Force Eurofighter Typhoons, which hold NATO’s Baltic Air Policing alert at Siauliai, carried out the intercept. Lithuanian President Gitanas Nauseda thanked NATO for its “swift response,” and Foreign Minister Kestutis Budrys said the intercept sent “a clear message that NATO stands vigilant, committed, and ready,” according to Deutsche Welle.

    Lithuania, which borders both Russia’s Kaliningrad exclave and Belarus, has reported frequent drone incursions since 2022, as have fellow NATO members Latvia and Estonia. The type of drone and its operator had not been established as of Tuesday.

    Why it matters

    Repeated drone incursions along NATO’s eastern flank are accelerating investment in layered counter-UAS detection below the fighter-jet tier, since scrambling manned aircraft for every low-cost drone is not a sustainable long-term response.

    Sources

  • Boon Edam to Demonstrate Biometric-Integrated Entrance Systems at GSX 2026

    Boon Edam to Demonstrate Biometric-Integrated Entrance Systems at GSX 2026

    Boon Edam Inc. plans to demonstrate entrance security systems integrated with biometric authentication technology at booth 2633 during GSX 2026 in Atlanta, according to Security Today. The lineup includes optical turnstiles, revolving doors and interlocking portals designed to detect unauthorized entry attempts such as tailgating or piggybacking.

    One demonstration pairs a Speedlane Compact optical turnstile with IDEMIA’s MorphoWave contactless fingerprint reader, which scans multiple fingerprints as a user passes a hand over the device. Another combines Boon Edam’s Circlelock interlocking portal with Alcatraz Rock facial authentication and an Iris ID reader, applying facial and iris recognition at different points of entry.

    A Tourlock 180 revolving door will be shown alongside the Alcatraz Rock X facial authentication system, supporting simultaneous multidirectional movement while monitoring for tailgating. Boon Edam will also display a mobile Turnlock full-height turnstile mounted on a movable steel base, featuring a BE Secure overhead system that uses artificial intelligence and measurement sensors to detect two people attempting to pass on a single credential, along with a Lifeline Speedlane Swing turnstile using a standard card reader.

    Why it matters

    Pairing physical entrance hardware with biometric readers at the point of passage, rather than at a separate door reader, is aimed squarely at tailgating, one of the most persistent gaps in access-control programs regardless of credential technology.

    Sources