Issue 34
Sysdig: AI agent carries out ransomware attack, but 'no human help' claim is disputed
Sysdig researchers say an AI agent called JADEPUFFER carried out a complete ransomware attack in 31 seconds by exploiting Langflow vulnerability CVE-2025-3248 (2026-07-06). Other reports say the "first" AI-driven attack still required human involvement.
Who is affected and how
- security leader
- The researchers say the attack chain runs at machine speed, so defence plans built on human reaction time need to be rethought.
- security operations
- Because an instance exploiting Langflow CVE-2025-3248 was involved, exposed Langflow deployments should be checked.
- developer
- For systems built on Langflow, the concrete tasks are patching CVE-2025-3248 and assessing how exposed instances are.
- journalist and researcher
- The "fully autonomous" label is disputed, so Sysdig's claim should be reported together with a precise description of the human role.
CISA adds Adobe ColdFusion and Oracle Agile PLM flaws to its Known Exploited Vulnerabilities catalog
On 2026-07-06, CISA added vulnerabilities in Adobe ColdFusion and Oracle Agile PLM to its Known Exploited Vulnerabilities catalog. Another report says the maximum-severity ColdFusion flaw is already…
security leaderThe listing sets a patching-deadline priority for risk management.security operationsCatalog listing signals exploitation, so patching ColdFusion and Oracle Agile PLM instances should be prioritised.research · globalGitLost: prompt injection flaw lets GitHub's AI agent leak private repository contents
Noma Labs says a prompt injection flaw called GitLost caused GitHub's AI agent to expose private repository data after a specially crafted public issue. The researchers say the safeguards were also…
security operationsThe permissions of agents connected to code repositories and their data exfiltration paths should be reviewed.developerAI agents with access to private repositories should be treated as attackable through public issues and other untrusted inputs.journalist and researcherNoma Labs' write-up is a concrete example of prompt injection affecting source code.jelentés · Egyesült KirályságUK agency: GPT-5.6 jailbreaks unlock dangerous cyber capabilities
A UK agency has found that jailbreaks of OpenAI's GPT-5.6 model unlock dangerous cyber capabilities (2026-07-10).
security leaderModel safeguards can be bypassed, so attacker capabilities must be factored into threat modelling.public-sector decision-makerThe result of government testing supports cybersecurity evaluation of frontier models.journalist and researcherThe agency's finding underlines the importance of measuring jailbreak resistance.vulnerability · United StatesCISA warns of active exploitation of SharePoint remote code execution flaw
On 2026-07-06, CISA warned that a SharePoint RCE vulnerability is being actively exploited.
security leaderBased on the agency warning, an emergency patching decision covering SharePoint is justified.security operationsGiven the active exploitation, patching SharePoint servers and checking their exposure is urgent.Hungary
More key stories of the week
M&S, Nationwide, Microsoft and ITV sign UK government's voluntary cyber resilience pledge
On 2026-07-06, DSIT and the NCSC launched the voluntary Cyber Resilience Pledge. Reports say there are more than 60 founding signatories, including M&S, Nationwide, Microsoft and ITV.
security leaderThe pledge may become a new benchmark for assessing security programmes and for supplier expectations.business leaderThe voluntary pledge is a public commitment to cyber resilience.regulation · European UnionEuropean Commission refers Ireland, Spain, France and Netherlands to EU Court of Justice over cybersecurity rules
On 2026-07-07, the European Commission referred Ireland, Spain, France and the Netherlands to the Court of Justice of the EU for failing to transpose the cybersecurity rules.
business leaderFor companies operating in the four countries, uncertainty over national rules affects compliance planning.public-sector decision-makerThe procedure shows that transposition delays can be enforced at EU level.regulation · European UnionEU presents action plan on cybersecurity risks and opportunities of advanced AI
On 2026-07-07 the European Commission presented an action plan to address the cybersecurity risks and opportunities of advanced artificial intelligence. A Hungarian-language report on it has also…
security leaderThe action plan signals the EU's direction on defensive and offensive uses of AI.public-sector decision-makerThe plan gives an EU framework for managing the cybersecurity risks of AI.bűnüldözés · ÁzsiaJapanese police arrest teenager over ChatGPT-assisted attack on anime and Bandai Channel accounts
Japanese police arrested a 15-year-old boy, according to reports of 2026-07-06, over a cyberattack that unsubscribed more than 46,000 anime accounts. Reports say the attack was assisted by ChatGPT…
security operationsDetecting and limiting mass automated actions against accounts is a defensive task for service providers.journalist and researcherThe case is an example of entry-level attacks supported by AI tools.incident · globalFake Paysafe and Skrill SDKs on npm and PyPI steal credentials
Fake Paysafe and Skrill SDK packages have appeared on npm and PyPI and steal credentials (2026-07-08).
security operationsIn build environments, dependency sources should be checked and credentials rotated if an affected package was pulled in.developerPackage names and publishers of payment SDKs should be verified before installation.research · globalHuntress analyzes vibe-coded malware used to map Active Directory environments
An analysis published on 2026-07-08 examines an AI-generated, vibe-coded malicious tool for Active Directory enumeration.
security operationsDetecting Active Directory enumeration is a defensive task regardless of where the tool came from.journalist and researcherThe analysis offers a concrete sample of AI-assisted malware development.vulnerability · globalMicrosoft patches RoguePlanet zero-day in Defender
According to reports on 2026-07-09, Microsoft took action to contain the zero-day threat known as RoguePlanet.
security leaderHandling the zero-day calls for a case-by-case risk assessment.security operationsFollowing Microsoft's action, the relevant updates and mitigations should be verified.incident · European UnionDutch police suspect Dutch hackers were involved in Odido data breach
Police said on 2026-07-10 that they suspect Dutch hackers were involved in the Odido data breach.
security leaderThe police suspicion shows the direction of investigations into service-provider data breaches.citizenThe breach puts the personal data of affected customers at risk, so notifications from the provider should be monitored.research · globalPrompt injection attacks can trick AI agents into making crypto payments
A report on 2026-07-06 says prompt injection attacks can be used to trick AI agents into making crypto payments.
security leaderFor agents able to move funds, deceptive inputs pose a direct risk of financial loss.developerAgents with payment authority should be protected with approval steps and value limits.bűnüldözés · Európai UnióSpain arrests suspect linked to pro-Russian hacktivist groups in operation with the FBI
Spain's Interior Ministry said the National Police, working with the FBI, arrested an alleged collaborator of the pro-Russian hacktivist groups CyberArmy of Russia Reborn (CARR) and Z-Pentest (7 July…
security leaderThe Western law-enforcement cooperation is an example of action against hacktivist groups.journalist and researcherThe official statement describes the suspect's link to the groups as alleged.attributionNew Iran-linked group targets Israeli government and IT sectors
According to Infosecurity Magazine, a new Iran-linked hacker group is targeting Israeli government and IT organisations (2026-07-06).
security leaderThe campaign against Israel's government and IT sectors is part of the regional threat picture.journalist and researcherAccording to the report, the Iran link is an attribution, not an established fact.research · globalHalluSquatting: AI coding assistants can be tricked into installing botnet malware
According to a report of 2026-07-08, an attack called HalluSquatting can trick AI coding assistants into installing botnet malware.
security operationsPackages suggested by AI that enter the development environment need to be verified.developerPackage names and dependencies suggested by an AI assistant must be checked before installation.In brief
Weekly email
One email a week with the main stories from the Agent Age editions. You confirm by email and can unsubscribe in one click.
Almost done: we sent you an email. Open it and press the confirm link. If it does not arrive within a few minutes, check your spam folder.
Please enter a valid email address and tick the consent box.
Something went wrong. Please try again later.