During a routine test that sent an AI model browsing random public websites, Anthropic’s Claude landed on Philadelphia’s unsolved homicides tip portal and submitted fabricated content claiming to have witnessed something related to a killing, leaving the name and contact fields blank. The submission hit the site on July 18. Anthropic says it discovered the entry on September 28 but did not notify Philadelphia police until October 7, a nine-day gap that police called unacceptable, and a total of 81 days since the original submission. 

The tip was filtered to spam and never reached investigators. The root cause was a permissions gap: the model had been told not to log in, create accounts, or make purchases, but submitting forms was never on that restricted list. Anthropic’s internal review reportedly found similar cases of an unreleased model navigating from practice government forms to live ones. With AI, what you ask it to leave out is as important as what you ask it to include.
https://www.gadgetreview.com/anthropics-ai-pretended-to-be-an-eyewitness-and-sent-philadelphia-police-a-fake-murder-tip

Did someone forward you this? Subscribe to this newsletter.

In other cybersecurity news:

The first example of significant real-world harm caused by unsafe AI would turn an already wary public further against the technology and its leaders, including Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, as well as President Trump for his reluctance to regulate it.
https://www.axios.com/2026/10/09/ai-companies-day-after-major-attack

The Register reports that AI systems are now fully capable of carrying out nightmare-style attacks against critical infrastructure, and that defenders are not ready. The piece describes AI agents moving robotic arms and making OT device operator screens lie about what equipment is doing.
https://threatintel.cc/2026/10/12/ai-systems-are-fully-capable.html

Thousands of cheap Android phones shipped with ad-fraud malware. Researchers have found malware preinstalled on thousands of cheap Android phones that can generate fraudulent ad revenue and potentially turn infected devices into parts of larger botnets.
https://therecord.media/cheap-androids-shipped-with-ad-fraud-malware

Tabung Haji Core Services And Systems Restored Following Cyberattack; Depositors’ Savings Confirmed Safe. Lembaga Tabung Haji has announced that its key systems and services have resumed operations following their temporary suspension last week. The financial institution also assured that depositors’ savings are confirmed safe. In a statement posted on its social media channels, Tabung Haji explained that it has restored these systems and functions after conducting comprehensive safety checks, as well as strengthening controls and implementing system recovery measures. The statement went on to assert that close monitoring and security enhancements will continue to ensure the recovered systems and services operate safely and smoothly.
https://www.lowyat.net/2026/406993/tabung-haji-systems-restored/

CyberLeek, the anonymous figure behind this summer’s wave of GTA 6 leaks, just raised the stakes considerably. After dropping 25 minutes of gameplay footage earlier last week, including nudity and dev menu access, they’re now threatening to release a complete, playable build of the game, start to finish. “A GTA 6 leaker has reportedly threatened to put out a complete, playable build of the game, start to finish.” reads Eurogamer.net. “The threat was posted on CyberLeek’s website after the leaker shared 25 minutes of footage from the game earlier this week.”
https://securityaffairs.com/200780/cyber-crime/gta-6-leaker-demands-30m-meme-coin-market-cap-to-release-full-game.html

A Russia-aligned cyber espionage group has steadily upgraded its MATCHBOIL downloader over two years, adding stronger obfuscation, sandbox checks and changes to its execution and payload persistence. In new research published on October 8, ESET documented MATCHBOIL versions compiled or observed between April 2024 and April 2026, finding that each version was more sophisticated than the one before it.
https://www.infosecurity-magazine.com/news/russia-aligned-uac-0099-evolves/

In July 2026, Zscaler ThreatLabz found a trojanized Terraform provider posing as an AWS plugin. Once Terraform loads it, the fake provider installs malware on macOS, Linux, or Windows. Researchers link the activity to suspected North Korean group TraderTraitor, but not with high confidence.
https://securityonline.info/tradertraitor-trojanized-terraform-provider

SonicWall has disclosed yet another critical flaw in one of its core products. CVE-2026-102255, rated 10 in severity, the highest possible on the Common Vulnerability Scoring System (CVSS), is a pre-authentication server-side request forgery (SSRF) vulnerability in the SMA1000 Appliance Work Place interface. An unintended access path could allow attackers to order the appliance to issue requests on their behalf and gain access to internal functions to perform unauthorized actions. At the same time, the cybersecurity company also disclosed three other vulnerabilities of lesser severity impacting the SMA1000, and “strongly advises” customers using the appliances to upgrade to the fixed release version. SonicWall said there is no evidence as yet that the critical vulnerability is being exploited in the wild.
https://www.csoonline.com/article/4232246/sonicwalls-latest-critical-flaw-indicates-a-security-pattern-not-another-one-off-bug.html

Lava Security discovered a high-severity vulnerability in the NVIDIA DCGM Exporter that allowed an unauthenticated attacker to trigger uncontrolled resource consumption on the GPU server, leading to denial of service and information disclosure. By exhausting the exporter’s resources, an attacker could crash the monitoring service, blind operators to GPU health and activity, and potentially slow workloads running on the same host.
https://lava.security/research/cve-2026-47483-nvidia-dcgm-exporter-vulnerability

South Korean and Japanese companies are racing to strengthen cyber defenses after a wave of attacks that cybersecurity experts say are highlighting how AI may be lowering the bar for criminals with limited technical skills. Nine South Korean banks and two mega-churches are investigating cyberattacks that may have involved AI tools, while Japanese companies including Daiwa Securities, SoftBank and the Lawson convenience store chain have been hit by a recent surge in cyber incidents.
https://www.nbcnews.com/world/asia/south-korea-japan-are-buffeted-hacks-ai-lowers-bar-cybercriminals-rcna602464

A YouTuber built a cop-watching ALPR system, prompting police to appear at his door.Flock Safety stores plate data in AWS Government Cloud, deleting it on a seven-day cycle.ALPR records can map where people live, work, worship, or seek medical care without suspicion.
https://www.gadgetreview.com/youtuber-builds-system-that-mimics-flock-to-spy-on-cops-then-the-cops-came-knocking

Russian bot traffic has dropped dramatically in the past few days after Ukraine knocked out Yandex’s biggest data centers, with a third attack happening last night. Cloudflare reports that bot traffic from the Yandex network has dropped to just 0.4% of the maximum number of HTTP requests recorded in the past two weeks. It initially dropped to around 57% on Oct. 8, which happened when Ukraine hit a Yandex data center housing two supercomputers in Sasovo, Russia, located around 195 miles to the southeast of the Kremlin. A second, bigger Yandex campus in Kaluga, about 220 miles to the west of Sasovo, was knocked offline yesterday.
https://www.tomshardware.com/tech-industry/data-centers/russian-disinformation-bot-traffic-nearly-eliminated-overnight-after-ukraine-drone-strikes-cripple-multiple-of-russias-google-data-centers-ukraine-intensifies-bombing-of-yandex-data-centers-in-response-to-moscows-strikes-on-kyiv-data-centers

A Flask cookie on developers.google.com was signed with a secret it was also carrying inside itself. Pulling that secret out turned a forged session into code execution across seven Google domains.
https://simonkoeck.com/writeups/google-devsite-flask-session-pickle-rce

Telegram Desktop hands clicked links to its own already-running instance over a local socket, as text, and never escapes the character it uses to separate commands. So a crafted link does not arrive as one instruction: it arrives as several.
https://beaksec.github.io/posts/telegram-desktop-one-click-account-takeover/

A huge trove of behind-the-scenes animation material from Cartoon Network’s history appears to have leaked online, and it’s the kind of stuff animation nerds rarely, if ever, get to see. The leak centers on a Google Drive folder containing production files from shows including The Amazing World of Gumball, Foster’s Home for Imaginary Friends, The Powerpuff Girls, The Looney Tunes Show, Duck Dodgers, and Garfield. Mashable was able to access the Drive and independently confirm its contents.
https://mashable.com/entertainment/massive-cartoon-network-leaks-exposes-hundreds-of-animation-files

Facebook could face billions of dollars in penalties after a New Mexico jury found it misled users about their personal data and other platform practices. The September 25 verdict found 43,899,725 violations of the state’s Unfair Practices Act, according to the New Mexico Department of Justice. The case covered Facebook’s privacy assurances, content moderation claims and response to the Cambridge Analytica scandal. That scandal involved personal information harvested from as many as 87 million Facebook users through a third-party app. Years later, the legal fallout continues for parent company Meta.
https://www.dexerto.com/entertainment/facebook-could-owe-billions-after-jury-finds-platform-responsible-for-43-million-privacy-content-violations-3414780/

Brandon Thomas, developer of ArtCraft, a suite of open-source creative tools meant to replicate Adobe applications for free, has now announced apps to replace Office tools such as Excel, PowerPoint, and Word, respectively, named GridCraft, DeckCraft, and WordCraft, that are created with Claude. Thomas has shared his new clones to ArtCraft GitHub, making them free for all to download.
https://www.tomshardware.com/software/microsoft-office/creator-of-free-adobe-clones-unveils-open-source-microsoft-office-excel-powerpoint-and-word-replications-autodesk-tools-and-other-subscription-based-services-may-come-next

Japan has declared a cybersecurity emergency, following a spate of attacks on its companies in recent weeks. Some experts are already pointing the finger at AI. Over 20 major firms, including convenience store chain Lawson and car rental operator Times Car, have reported attacks, according to The Financial Times, leading to the loss of tens of millions of pieces of customer data. This included 1.67 million customer email addresses from the booking system of JR East, the country’s largest rail operator.
https://www.pcmag.com/news/japan-declares-cybersecurity-emergency-experts-nod-to-ai

A North Carolina man siphoned millions in royalties from Spotify and its rivals with songs no human wrote, played to listeners who never existed. Every one of those phantom plays skimmed from the same pot that pays working musicians, which is why platforms spent this year swatting at AI clutter, from anime waifus hijacking jazz legends’ profiles to badges for synthetic acts.
https://www.dexerto.com/entertainment/man-sentenced-to-prison-after-pocketing-8-million-from-ai-songs-streamed-by-bots-3416335/

At least three accounts at the company linked to a major breach of Denmark’s CPR register reportedly used the password “123456”, including an administrator account. At least three user accounts at the Funen-based IT company Pays used the password “123456” when hackers gained access to Denmark’s CPR register, Politiken reports.
https://cphpost.dk/2026-10-10/news/round-up/123456-password-used-in-massive-danish-cpr-data-breach/

Daiichi Kosho, a major Japanese entertainment system maker, disclosed that a malware infection at its contractor, Nippon Columbia, exposed more than 8.7 million customer and employee records. The company is the largest karaoke maker in Japan, operating 521 karaoke venues nationwide, including the Big Echo chain, one of the most popular karaoke box chains in the country.
https://www.bleepingcomputer.com/news/security/nippon-columbia-malware-incident-exposes-86-million-karaoke-fan-records/

CISA published advisory ICSA-26-272-06 detailing a critical vulnerability in MikroTik RouterOS. A single crafted HTTP request allows unauthenticated remote code execution as root via an integer underflow in the web management service. Over 367,000 web interfaces are exposed online per Censys, against a backdrop of 2.8 million total MikroTik devices detected by Shadowserver Foundation. MikroTik has not issued a dedicated security advisory; CISA identifies RouterOS 7.24 as the fixed version, leaving the status of intermediate 7.23.x releases unclear.
https://deafnews.it/en/news/news/cve-2026-84411-pre-auth-integer-underflow-threatens-367000-exposed-routeros-interfaces

A catastrophic ransomware attack has severely crippled a significant portion of the Japanese cloud platform, IDCF Cloud, a critical infrastructure component upon which hundreds of corporate enterprises and government agencies rely. Currently, 495 organizations are confirmed victims. IDC Frontier, the cloud operator, has ominously warned that independent data recovery on the compromised servers is highly improbable. Consequently, clients lacking isolated, external backups face the grim prospect of permanent, irreversible data loss.
https://meterpreter.org/idcf-cloud-ransomware-attack-japan