Worldwide, weighted to the United States.
Cybersecurity
10 POINTS
Authorized researchers at Hacktron used Claude Opus 4.8 and Opus 5 to build a working exploit chaining an unpatched flaw in the libheif image library with excessive permissions in an OpenAI sign-in token, gaining access to employee ChatGPT and Codex accounts and limited reads of private-repository metadata and commits on an internal GitHub repository. OpenAI confirmed a fix within about 14 hours of the researchers' report through Bugcrowd, narrowed the sign-in token's permissions, revoked affected sessions, and paid a $6,500 bounty; the underlying libheif flaw was patched within two days.
SecurityWeek · September 18, 2026
Why it matters: A frontier AI lab's own internal systems were reached using an exploit that a rival lab's model helped construct, showing that AI-assisted offensive research is now fast enough to find real production flaws inside the companies building the models.
Watchpoint: The researchers say they had only theoretical, unverified access to Slack and other internal services beyond GitHub and the sign-in tokens, so the true blast radius of the flaw was never fully tested.
- Security researchers at Hacktron, working through OpenAI's Bugcrowd bounty program, disclosed the findings September 18, 2026.
- They used Anthropic's Claude Opus 4.8 and Opus 5 models to help build a working exploit chain.
- The exploit combined an unpatched flaw in the libheif image-processing library with a flawed OpenAI sign-in token.
- The sign-in token flaw granted excessive permissions, including broader API access than intended.
- Using the chain, researchers accessed employee ChatGPT and Codex accounts.
- They also obtained limited reads of private-repository metadata and commit history on an internal OpenAI GitHub repository.
- OpenAI confirmed a fix to the sign-in token issue about 14 hours after the researchers reported it via Bugcrowd.
- OpenAI narrowed the permissions on Community sign-in tokens and revoked the affected tokens and sessions.
- The libheif library flaw, reported separately to its maintainers, was patched within about two days.
- OpenAI paid a $6,500 bounty for the account-takeover finding and thanked the researchers for responsible disclosure.
10-point recap written from the linked reporting. The article itself is the source.
Cybersecurity
10 POINTS
Screenshot-sharing service Gyazo, operated by Japanese company Helpfeel, disclosed that a flaw in its image-upload server let an attacker run arbitrary commands and access its database, exposing 23.62 million user records and 490 million image metadata records. Helpfeel detected the intrusion on the evening of September 11, blocked access and patched the vulnerability by the next morning, reported the incident to Japan's Personal Information Protection Commission on September 15, and published its public notice on September 16. The company said no payment card information was exposed and has not disclosed the specific vulnerability type.
BleepingComputer / The Hacker News · September 16, 2026
Why it matters: Nearly 500 million image metadata records from a single screenshot tool is a reminder that services storing years of casual uploads accumulate a scale of exposure that dwarfs their apparent size, long after most users have forgotten they ever signed up.
Watchpoint: Helpfeel has not disclosed the specific vulnerability type that let the attacker run arbitrary commands, so it is not yet possible to say whether other Helpfeel-operated services share the same flaw.
- Gyazo, a screenshot-sharing service operated by Japanese company Helpfeel, disclosed a data breach on September 16, 2026.
- The breach exposed 23.62 million user records, according to Helpfeel's own disclosure.
- It also exposed roughly 490 million image metadata records tied to uploads on the platform.
- The cause was a vulnerability in Gyazo's image-upload server that let an attacker run arbitrary commands.
- That access let the attacker reach and extract data from Helpfeel's backend database.
- Helpfeel detected the intrusion on the evening of September 11, 2026, Japan time.
- The company blocked the access routes and patched the vulnerability by the morning of September 12.
- Helpfeel reported the incident to Japan's Personal Information Protection Commission on September 15.
- The company said no payment information, including credit card numbers, was exposed in the breach.
- Helpfeel has not specified the exact type of vulnerability that was exploited.
10-point recap written from the linked reporting. The article itself is the source.
Cybersecurity
10 POINTS
Cisco disclosed CVE-2026-76460, a maximum CVSS 10.0 authentication-bypass flaw in Cisco Identity Services Engine and ISE Passive Identity Connector, caused by insufficient authentication control on an API endpoint that lets an attacker bypass the web management interface and execute commands with root privileges. Cisco confirmed active exploitation and released patched versions across the 3.1 through 3.5 branches, and is advising customers to check access logs for a specific suspicious username pattern and to re-image and restore from backup any node found to be compromised.
The Hacker News · September 17, 2026
Why it matters: A maximum-severity, actively exploited flaw in the identity and access-control platform many enterprises use to gate their own networks is the kind of bug that, left unpatched, undermines every other security control sitting behind it.
Watchpoint: Cisco's guidance notes that a successful attacker can hide evidence of compromise using the same root access, so organizations that only check logs without re-imaging suspect nodes may wrongly conclude they were not affected.
- Cisco disclosed the vulnerability, tracked as CVE-2026-76460, on September 17, 2026.
- It carries the maximum CVSS score of 10.0, Cisco's highest severity rating.
- The flaw affects Cisco Identity Services Engine (ISE) and ISE Passive Identity Connector (ISE-PIC).
- The root cause is insufficient authentication control on an API endpoint in the product.
- Exploiting it lets an attacker bypass the web-based management interface entirely.
- A successful exploit grants command execution with root privileges on the affected system.
- Cisco confirmed it is aware of active exploitation of the vulnerability in the wild.
- Patched versions are available across the 3.1, 3.2, 3.3, 3.4 and 3.5 branches.
- Cisco advises checking access logs for a specific suspicious username pattern tied to exploitation attempts.
- For any node suspected of compromise, Cisco recommends re-imaging and restoring from a clean configuration backup.
10-point recap written from the linked reporting. The article itself is the source.
Cybersecurity
10 POINTS
Check Point disclosed CVE-2026-91843, a 9.8-severity stack-overflow flaw in its Security Management and Log Servers triggered by login requests containing extremely long usernames, which lets an unauthenticated attacker execute code with root privileges over the network if a system's Trusted Clients setting is not restricted to known hosts. Check Point said it has no reports of the flaw being exploited in the wild and released a LivePatch fix under advisory sk1000155, while urging customers to verify their Trusted Clients access is limited rather than open to any IP address.
The Hacker News · September 17, 2026
Why it matters: A root-level, pre-authentication flaw in the servers that manage an organization's firewall policy is a single point of failure for network security teams, and the fact that it is unexploited today is a matter of timing, not design.
Watchpoint: The flaw is only exploitable through the Trusted Clients setting, so its real-world risk depends entirely on how many Check Point customers have left that setting open to any IP address rather than a restricted list.
- Check Point disclosed the vulnerability, tracked as CVE-2026-91843, with a CVSS score of 9.8.
- It affects Check Point Security Management and Log Servers across versions R80 through R82.20.
- The root cause is a stack overflow triggered during the login process before authentication completes.
- The overflow is triggered specifically by login requests containing extremely long usernames.
- A successful attacker can execute code with root privileges on the affected server over the network.
- Exploitation is possible only through the product's Trusted Clients setting rather than an unrestricted attack surface.
- Check Point said it has received no reports of real-world exploitation as of the disclosure date.
- CISA's own assessment likewise recorded no known exploitation at the time of publication.
- Check Point released a LivePatch fix for the issue under advisory sk1000155.
- The company is urging customers to confirm Trusted Clients access is restricted to known hosts, not any IP address.
10-point recap written from the linked reporting. The article itself is the source.
Cybersecurity
10 POINTS
Attackers obtained a long-lived Cloudflare API key with full account permissions that was hardcoded in marketing platform Brevo's application source code, used it to create a malicious Cloudflare Worker, and injected fake browser-verification pages and malicious scripts into roughly 100,000 customer websites embedding Brevo's widgets and SDKs, according to security firm Sansec. The scripts directed Windows visitors to run attacker commands and attempted to install a backdoor plugin disguised as a 'Web Media Optimizer' plugin on WordPress sites with logged-in administrators; the compromise ran for about five and a half hours on September 14 before Brevo's security headers and integrity checks were restored.
BleepingComputer · September 17, 2026
Why it matters: A single hardcoded API key inside one marketing vendor's own codebase turned into a five-and-a-half-hour supply-chain attack across 100,000 unrelated websites, which is exactly the blast radius that makes third-party embedded scripts a bigger risk than most site owners realize.
Watchpoint: Brevo has not said how long the exposed Cloudflare key had been hardcoded in its source before it was used in this attack, so affected customers cannot yet tell whether earlier, undetected abuse also occurred.
- Security firm Sansec identified the supply-chain attack against marketing platform Brevo, reported September 17, 2026.
- Attackers had obtained a long-lived Cloudflare API key with full account permissions hardcoded in Brevo's source code.
- They used that key to create a malicious Cloudflare Worker modifying content at the CDN edge.
- The Worker injected fake Cloudflare verification pages and malicious scripts into Brevo's embedded website components.
- Roughly 100,000 websites using Brevo's forms, widgets or SDK loaders were potentially affected.
- On Windows, the fake verification pages tried to trick users into running attacker-supplied commands.
- On WordPress sites with a logged-in administrator, the script attempted to install a backdoor disguised as a plugin.
- The malicious Worker also stripped security headers like Content-Security-Policy to evade standard integrity checks.
- The compromise window ran from roughly 16:07 to 20:30 UTC on September 14, 2026, about five and a half hours.
- Brevo has not disclosed how long the exposed API key had existed in its codebase before this incident.
10-point recap written from the linked reporting. The article itself is the source.
Cybersecurity
10 POINTS
The FBI, the U.S. Defense Department, and police in Japan, Australia and Germany jointly attributed a campaign called WaterPlum to North Korean hackers who posed as AI and blockchain companies recruiting job seekers, tricking applicants into downloading malware disguised as interview materials. Between December 2025 and July 2026 the campaign infected at least 30,000 devices across 100 countries, compromised roughly 7,000 cryptocurrency wallets and stole more than $10.5 million, using malware families including BeaverTail, InvisibleFerret, OtterCookie, OtterCandy and StoatWaffle, with some infections kept dormant in hopes targets later joined tech firms and could enable deeper corporate network access.
The Record · September 18, 2026
Why it matters: A years-long, multi-country fake-hiring operation that plants persistent access on job seekers' machines in hopes they later land a job at a real tech company shows North Korea treating the tech hiring pipeline itself as a long-horizon espionage target, not just a quick cryptocurrency theft scheme.
Watchpoint: Investigators say some WaterPlum infections were kept dormant specifically in hopes the victim later joined a legitimate tech company, meaning the campaign's full damage may not be visible until infected job seekers are hired months or years from now.
- The FBI and international law-enforcement partners attributed the WaterPlum campaign to North Korean hackers, reported September 18, 2026.
- Partner agencies included the U.S. Defense Department and police in Japan, Australia and Germany.
- The campaign ran from December 2025 through July 2026.
- Attackers posed as AI and blockchain companies to recruit real job seekers.
- Applicants were tricked into downloading malware disguised as interview or onboarding materials.
- At least 30,000 devices were infected across 100 countries during the campaign.
- Roughly 7,000 cryptocurrency wallets were compromised as part of the operation.
- Investigators say more than $10.5 million was stolen through the campaign.
- Malware families used include BeaverTail, InvisibleFerret, OtterCookie, OtterCandy and StoatWaffle.
- Some infections were kept dormant on the chance a victim later joined a tech firm, enabling deeper corporate access.
10-point recap written from the linked reporting. The article itself is the source.
Cybersecurity
10 POINTS
The Department of Justice, the FBI's Anchorage field office, the Royal Canadian Mounted Police and the U.S. Attorney's Office for the District of Alaska seized the domains behind NightmareStresser, a DDoS-for-hire platform assessed to have facilitated hundreds of thousands of actual or attempted denial-of-service attacks worldwide since 2022, with a prior report citing over 566,000 registered users. The seizure is part of Operation PowerOFF, a broader international law-enforcement effort against DDoS-for-hire infrastructure that has now charged twelve defendants tied to such services.
The Hacker News · September 17, 2026
Why it matters: A platform with over half a million registered users staying online since 2022 shows how cheaply DDoS attacks have been commoditized, and a domain seizure alone rarely stops operators from simply standing up new infrastructure elsewhere.
Watchpoint: The announcement does not name any arrests specifically tied to NightmareStresser's operators, only domain seizures, so it remains unclear whether the people running the service face prosecution or can relaunch it under a new name.
- US and Canadian authorities announced the seizure of NightmareStresser's domains on September 17, 2026.
- Agencies involved included the Department of Justice, the FBI's Anchorage field office and the RCMP.
- The U.S. Attorney's Office for the District of Alaska also participated in the action.
- NightmareStresser is assessed to have facilitated hundreds of thousands of actual or attempted DDoS attacks since 2022.
- A prior report had cited more than 566,000 registered users on the service.
- The two seized domains were nightmare-stresser[.]com and nightmarestresser[.]org.
- The action is part of Operation PowerOFF, a broader international campaign against DDoS-for-hire infrastructure.
- Operation PowerOFF has now resulted in charges against twelve defendants who facilitated DDoS-for-hire services.
- The announcement does not detail any arrests specifically tied to NightmareStresser's own operators.
- DDoS-for-hire platforms let paying customers rent attack capacity against a target with little technical skill required.
10-point recap written from the linked reporting. The article itself is the source.
Cybersecurity
10 POINTS
A critical flaw in Orkes Conductor, an open-source workflow orchestration platform, lets unauthenticated attackers achieve remote code execution by submitting malicious JavaScript or Python expressions to the workflow API on servers running unsandboxed GraalVM evaluators. Tracked as CVE-2026-58138 with a CVSS score of 9.8, the flaw affects versions 3.21.21 up to 3.30.2; Fortinet said it blocked 1,290 exploitation attempts in a single 24-hour period, a 132% jump in daily activity, with attack traffic traced mainly to Germany, Hong Kong, Indonesia, the UAE and India. Orkes has released version 3.30.2 to fix the issue.
The Hacker News · September 19, 2026
Why it matters: Workflow-orchestration platforms sit at the center of a company's automation pipelines with broad system access by design, so a pre-authentication remote-code-execution bug in one is a direct route into whatever that automation touches.
Watchpoint: Fortinet's blocked-attempt count only reflects traffic it can see through its own products, so the true scale of exploitation against Orkes Conductor deployments that are not behind Fortinet infrastructure is unknown.
- The Orkes Conductor vulnerability, tracked as CVE-2026-58138, was reported under active exploitation on September 19, 2026.
- It carries a CVSS v3.1 score of 9.8 and a CVSS v4 score of 9.3.
- Orkes Conductor is an open-source workflow orchestration platform used to automate business processes.
- The flaw allows unauthenticated remote code execution through the platform's workflow API.
- Exploitation works by submitting malicious JavaScript or Python expressions to that API.
- The root cause is unsandboxed GraalVM evaluators configured with full host access.
- Affected versions range from 3.21.21 up to, but not including, the patched 3.30.2 release.
- Fortinet reported blocking 1,290 exploitation attempts in a single 24-hour period, a 132% increase in daily activity.
- Roughly 7,000 attempts were blocked in total between September 2 and September 9, 2026.
- Attack traffic was traced primarily to Germany, Hong Kong, Indonesia, the UAE and India.
10-point recap written from the linked reporting. The article itself is the source.
