Issue 53 · Nov 16–22, 2026Agent Age News magazine · Other editions · Search · Method · HU / ENG

Incidents and influence attempts in the age of agents

Issue 43

Sep 7–13, 2026 · 81 stories · 8 of them from Hungary and the regionPage last updated: Nov 21, 2026
Viewpoint

report · global

Anthropic: AI lets smaller actors run state-level hacking campaigns

On 10 September 2026, Anthropic published its most detailed threat intelligence report to date on the misuse of Claude. The company says it disrupted every operation in the report and tightened defences against AI-based phishing and data theft.

Sep 10, 2026 · 12 sources

Who is affected and how

security leader
The report says AI appears in attacks at varying levels of autonomy, which affects threat modelling.
security operations
Anthropic says attacker sophistication no longer depends on the size of the actor, so detection assumptions need rethinking.
journalist and researcher
The report gives details of cases ranging from cyberattacks to influence operations, surveillance, biology and weapons building.
regulation · European Union

EU Cyber Resilience Act's 24-hour vulnerability reporting obligation takes effect

The reporting obligations of the EU Cyber Resilience Act came into force on 11 September 2026. Hardware and software manufacturers must report actively exploited vulnerabilities within 24 hours.

security leaderThe PSIRT and vulnerability disclosure process become an auditable expectation.developerProduct security and vulnerability management processes must be aligned with the reporting deadlines.business leaderManufacturers placing products on the EU market face greater legal liability because of the reporting obligation.
incident · European Union

NATO allies reportedly foil Russian plot to sabotage undersea cables

NATO allies have prevented a Russian plot to sabotage undersea cables, according to an exclusive report. The news was published on 10 September 2026 and picked up by many outlets.

security leaderPhysical threats to undersea communications infrastructure are part of critical-infrastructure risk assessment.public-sector decision-makerIntelligence cooperation among allies underpins decisions on cable protection.
vulnerability · United States

CISA adds actively exploited MikroTik RouterOS authentication flaw to KEV catalog

Around 10–11 September 2026, CISA added a MikroTik RouterOS authentication bypass flaw to its catalog of actively exploited vulnerabilities. Details are scarce.

security leaderA KEV listing triggers a US federal remediation requirement.security operationsMikroTik devices must be patched urgently or their access restricted.
attribution · global

Russian state-sponsored hackers reportedly used Claude to rebuild detected malware

According to reports of 11 September 2026, Russian state-sponsored hackers used Claude to rewrite malware after it was detected, and deployed it against Ukraine. The reports are based on an Anthropic…

security leaderState actors' use of AI is also a factor in defence planning for the region near Ukraine.security operationsRapidly rebuilding detected malware with AI can shorten the period in which signature-based defences remain effective.journalist and researcherThe attribution rests on Anthropic's claim; the excerpts contain no independent confirmation.

Hungary

incidentRevolut data breach: customer data obtained through fake government requestssecurity leaderData disclosure triggered by forged official requests shows that the process for authenticating government requests needs review.business leaderFor a financial services provider, how official requests are verified and how customers are notified is a direct legal and trust risk.citizenPassport copies and transaction data raise the risk of phishing and identity theft; any contact from the company should be verified.incidentRansomware emails impersonating the Hungarian prosecution service target organisationssecurity operationsEmails with attachments that appear to come from the prosecution service merit a separate filtering and reporting rule.business leaderOpening emails that cite an authority can cause operational shutdowns, so warning staff is advisable.public-sector decision-makerMisuse of an authority's name also undermines trust in the authenticity of official correspondence.citizenThe attachment of an email posing as an authority should not be opened without verification.fraudPhishing wave targets MVM customers againsecurity leaderA recurring campaign impersonating a utility provider, affecting brand-protection and customer-notification processes.citizenLinks and data requests in messages sent in MVM's name should be checked through the official channel.fraudPhishers and fraudsters now use AI, making fake content more convincingsecurity leaderUser-awareness training materials need to be updated to cover AI-generated scams.citizenBecause messages are linguistically flawless and look authentic, spotting scams by their language errors is less reliable.attributionRussian hackers may use fake CAPTCHA to breach Ukrainian government computerssecurity leaderThe fake-CAPTCHA lure (a ClickFix-style technique) makes protecting user endpoints urgent.public-sector decision-makerThe targeting of government networks warrants a review of public administration defences in the region.regulationA magyar Btk.-ban nincs deepfake-paragrafuspublic-sector decision-makerAccording to the article, the criminal-law treatment of deepfake abuse is a regulatory question.citizenAccording to the article, legal protection for deepfake victims is not tied to a dedicated statutory offence.regulationPolish consumer watchdog UOKiK orders banks to re-examine scam reportsbusiness leaderBanks must prepare to handle scam reports again and to make new refund decisions.citizenIn Poland, scam victims can ask again for their cases to be reviewed.fraudNew phishing wave aims to steal bank card details in HungarycitizenBank card details should not be entered in a form reached through a link.

More key stories of the week

research · global

Autonomous AI agents compromise thousands of credentials in under six hours

According to reports of 8 September 2026, autonomous AI agents compromised thousands of credentials in less than six hours. Help Net Security says attackers use agents to automate both vulnerability…

security leaderAttacks that scale without human intervention require rethinking credential management and detection speed.security operationsBecause credential theft is fast and automated, rotating secrets and monitoring for anomalies are urgent.
vulnerability · global

Microsoft's September Patch Tuesday fixes a record 973–974 flaws, including two exploited zero-days

Microsoft released its September patches on Sept. 8, 2026. Reports say 973 or 974 vulnerabilities were fixed, including two zero-days already being exploited.

security leaderThe exceptionally high patch volume strains testing and release processes.security operationsPatches for the two exploited zero-days should be deployed first.
incident · global

OpenAI agents' RubyGems attack came two months before Hugging Face incident

Reports say a RubyGems attack linked to OpenAI agents came two months before the Hugging Face breach involving OpenAI. The story gained further details between 9 and 11 September 2026.

security leaderLinked incidents across the AI supply chain point to vendor risk.developerThe chain between package registries and model platforms makes it necessary to check AI development dependencies.
incident · European Union

Rhysida publishes Berlin government data after €2m extortion demand is refused

The Rhysida ransomware group published Berlin government data after the €2m ransom demand went unpaid. According to Hungarian and German reports, state secrets and critical-infrastructure data may…

security leaderFor a public body, the task is to quickly assess which categories of published data were exposed and to protect the affected systems.public-sector decision-makerReports say state secrets and critical-infrastructure data may be affected, which calls for crisis management and a review of data-handling practices.citizenReports say the leak of personal data may affect Berlin residents.
vulnerability · United States

CISA flags exploited Cisco, Citrix and Fortinet flaws, sets Sept. 12 federal patch deadline

According to reports dated Sept. 10, 2026, CISA flagged exploited Cisco, Citrix and Fortinet flaws and set a Sept. 12 patch deadline for federal agencies. Another report says it added seven exploited…

security leaderThe KEV catalog is a prioritization signal for organizations outside the United States as well.security operationsPatching the listed Cisco, Citrix and Fortinet products is an immediate task.
vulnerability · global

Actively exploited N-able N-central zero-day allows unauthenticated remote code execution

The N-able N-central zero-day vulnerability CVE-2026-86218 is being actively exploited. The flaw allows unauthenticated remote code execution and authentication bypass; the news is dated Sept. 7…

security leaderCompromised remote management tools can affect multiple customers.security operationsMSPs running N-central, and their customers, need to patch urgently.
incident · global

Hundreds of AI agents compromise PaperCut servers, GreyNoise says

GreyNoise has identified an AI-driven intrusion campaign that, according to the articles, compromised 440 PaperCut servers, and by another report 395 organisations. The company says a…

security leaderAttacks automated with AI agents speed up exploitation relative to the patching window.security operationsPatching the PaperCut vulnerabilities is urgent, as the campaign uses agents to scan and attack at scale.
incident · European Union

Revolut confirms customer data breach through fake government requests

Revolut has confirmed that customer data leaked through fake government requests. The report is dated 12 September 2026.

security leaderThe verification process for government requests is an attack surface.citizenRevolut customers' data may have been exposed, and the risk of phishing is rising.
vulnerability · global

September Patch Tuesday: two zero-days and nearly 1,000 fixes for Windows

Microsoft's September Patch Tuesday package closes two zero-day vulnerabilities. According to the roundup, Adobe and Cisco released critical updates, and SAP fixed a critical NetWeaver flaw.

security leaderAccording to the roundup, critical updates from several vendors arrived at once, which calls for a patch-prioritisation decision.security operationsPrioritise the zero-day fixes, the flaws rated wormable, and the SAP NetWeaver update.
report · global

Google GTIG: AI agents carried out credential theft in under six hours

Google's Threat Intelligence Group (GTIG) says AI agents carried out credential theft in under six hours.

security leaderAccording to GTIG, agents have accelerated the pace of attacks, which warrants a review of response times.security operationsDetection and credential-revocation processes must work within hours.
incident · global

Attack on bitcoin-based network drains $320 million

Reports say $320 million was taken from a crypto network. The news dates from 7 September 2026.

business leaderSecurity flaws in crypto platforms cause direct financial losses.citizenThe risk of losses for crypto asset holders is high.
regulation · United Kingdom

UK peers question why cyber bill leaves executives without personal liability

The Lords committee stage of the UK Cyber Security and Resilience Bill ended on 9 September 2026. Peers asked why executives are left out of personal liability.

business leaderThe debate is about whether executives bear personal responsibility for cyber defence.public-sector decision-makerThe bill shapes the resilience requirements that apply to UK organisations.

In brief

incidentCyberattack halts Boston Scientific manufacturing and shipmentssecurity leaderSegmenting manufacturing systems and having a business continuity plan are urgent matters.business leaderDisruption to the medical device supply chain is a business and patient-care risk.incidentIDScan.net breach exposes driver's license data, with reports of more than 150 million recordsbusiness leaderIdentity-verification providers face growing responsibility for data protection.citizenDriver's license data can be used for identity theft.incidentAnthropic discloses another AI hacking incident missed in earlier reviewsecurity leaderGaps in AI providers' misuse detection bear on vendor risk management.journalist and researcherThe after-the-fact discovery of detection gaps raises questions about the reliability of vendor reports.reportCoinGecko: crypto platforms lost $3.63 billion to cyberattacks despite security auditssecurity leaderAccording to the report, an audit alone is not sufficient protection.business leaderRisk management that relies on security audits should be reviewed.incidentBerlin investigates new data leak after hackers publish stolen government login credentialssecurity operationsLeaked credentials must be revoked and logs analyzed.public-sector decision-makerLeaked credentials for government systems call for a review of access management.regulationFBI and EPA warn of cyberattacks on water systems as Texas becomes federal pilot sitesecurity operationsWater utility operators face urgent pressure to strengthen defences, given the official warnings and the need for visibility into OT networks.public-sector decision-makerThe federal pilot programme in Texas shows an organisational model for protecting critical infrastructure.regulationEU cybersecurity agency ENISA granted access to Anthropic's Mythos 5 model, Commission sayspublic-sector decision-makerDirect model access for the EU agency strengthens oversight of AI risks.journalist and researcherThe terms and scope of the access are not clear from the announcement.incidentTrezor data breach now affects 81,000 customerscitizenHardware wallet customers' data is affected, and the risk of targeted phishing is rising.attributionNorth Korea recruits abroad for its scheme to infiltrate US companiessecurity leaderInsider threat is a significant attack vector.business leaderIn remote-hire processes, identity verification is part of a company's internal risk.incident220 million traveler records exposed: nine years of airline datacitizenThe leak of passport and passenger data carries a risk of identity theft.regulationSen. Hawley opens investigation into OpenAI over hacking and AI risksbusiness leaderCompanies that contract with AI providers need to factor vendor risk into their planning because of the investigation.public-sector decision-makerThe investigation is an example of US congressional action on the oversight of AI providers.vulnerabilityCisco confirms Secure FMC flaw CVE-2026-20079 is exploited in attackssecurity operationsPatching Secure FMC instances and reviewing their logs is urgent.vulnerabilityStyleSmuggler: unpatched zero-day in Magento and Adobe Commerce actively exploitedsecurity operationsWithout a patch, compensating controls are needed on web shops.business leaderWeb shops' payment data may be at risk.incidentCyberattack disrupts operations at Malaysia's Port of Tanjung Pelepas, Hapag-Lloyd sayssecurity leaderCyber defence and business-continuity planning for logistics hubs is a priority.business leaderThe port disruption affects supply chains.incidentShinyHunters claims Carhartt data breach affecting nearly 13 million peoplecitizenAffected people's data may have been exposed, and the risk of misuse of personal data is rising.incidentBigBear Microsoft 365 phishing service bypassed MFA at 258 organizationssecurity leaderReviewing access logs of Microsoft 365 tenants is warranted.security operationsTraditional MFA does not protect against such phishing proxies, so more resistant methods are needed.fraudAttackers use passkey-themed scams to hijack Microsoft 365 accountssecurity operationsDisabling or restricting the device-code flow reduces the risk.citizenEmployees must be prepared for approaches that pose as the helpdesk.researchCheck Point: ChatGPT hidden channel let attackers read victims' Gmailsecurity leaderLinking AI assistants to email accounts opens a new data-leak channel.citizenData of users who linked a Gmail account to ChatGPT may have been affected.incidentData of 2.5 million student loan borrowers exposed via Nelnet servicingcitizenStudent loan borrowers' data has been exposed, raising the risk of fraud.vulnerabilityCISA adds a Chromium V8 flaw to its Known Exploited Vulnerabilities catalogsecurity operationsBrowser updates need to be rolled out quickly, and the patch process aligned to the two-week cadence.public-sector decision-makerA KEV listing sets federal-agency remediation deadlines for US organisations.influence operationRussian disinformation networks threaten European elections, official warnspublic-sector decision-makerAccording to the warning, election integrity must be protected through network monitoring and coordination.journalist and researcherThe excerpts contain no measured reach, so the claim of impact cannot be substantiated.citizenChecking the source of political content ahead of elections is important.incidentHackers are stealing Claude tokens from subscriberssecurity operationsOversight of organisational AI accounts and tokens is part of access management.citizenStealing tokens for AI services can allow accounts and paid subscriptions to be abused.vulnerabilityWiz reports in-the-wild exploitation of three JFrog Artifactory vulnerabilitiessecurity operationsArtifactory instances need to be patched and their logs reviewed.developerA compromised artifact repository can affect the entire software supply chain.researchFour spy groups used the same Chrome and Windows exploit kit within a weeksecurity operationsPatching Chrome and Windows takes priority.journalist and researcherThe shared toolkit complicates attribution conclusions.vulnerabilityCVE-2026-82533: DeepSeek harness flaw lets AI agents escape their sandboxsecurity leaderAgent isolation cannot be considered guaranteed.developerBecause agents can escape sandboxes, the framework must be updated and the runtime environment hardened.incidentKnown Shai-Hulud payload reappears on npm after 111 days and evades scannerssecurity operationsRotating build environments and secrets is warranted.developernpm dependencies need to be checked in addition to relying on scanners.attributionReport: Kimsuky used AI coding agents to create decoyssecurity leaderAI-made decoys call for stronger phishing defenses.journalist and researcherThe use of AI agents in state actors' toolkits is a subject of research.reportRansomware attacks hit record high in August, with Qilin leadingsecurity leaderThe ransomware threat level continues to rise.researchBlueMoon exploit kit turns Chrome and Windows flaws into attackssecurity leaderThe exploit kit points to chained exploitation of browser and operating-system flaws.security operationsDelays in patching Chrome and Windows are a direct risk because of the exploit kit.incidentFake IT calls target executives in Microsoft 365 data theft and extortion attackssecurity operationsStrengthen the approval processes for Microsoft 365 access and helpdesk authentication.business leaderExecutives are prime targets of fake IT calls, so dedicated awareness training is needed.researchNew MCP specification could turn a planted prompt into stolen credentialssecurity leaderAgents' tool access and tokens must also be protected against instructions arriving through prompts.developerThose working with MCP servers and agents need to redesign their handling of prompt injection and credentials.regulationChina's Supreme People's Court sets first legal rules on AI deepfakes and falsehoodspublic-sector decision-makerThe court guidance is an example of a regulatory approach to deepfake liability.journalist and researcherThe content of the rules is not known from the excerpts; details are in the source.vulnerabilityNew zero-day threatens Magento and Adobe Commerce online storessecurity operationsReview the exposure of Magento/Adobe Commerce instances and the compensating controls in place.business leaderFor an online store, exploitation means risk to card data and revenue.caseAmatera stealer spreads via fake CAPTCHA built on ClickFix techniquesecurity operationsCheck endpoint protection and script-execution restrictions against commands pasted by users.citizenDo not follow a page disguised as a CAPTCHA that asks you to copy and run a command.vulnerabilityMikroTik RouterOS authentication bypass flaw actively exploited, Hungarian cyber agency sayssecurity operationsCheck that RouterOS devices are updated and that access to the management interface is restricted.business leaderA flaw in a network edge device can affect the protection of the entire internal network.reportNCSC warns shadow AI creates new security riskssecurity leaderAccording to the NCSC's warning, organisations should assess which AI tools are being used without authorisation.business leaderThe data-handling and compliance risks of shadow AI call for management-level policy.regulationEU law: entering a code does not mean authorising a transfer, banks must repaybusiness leaderFinancial institutions need to review their fraud liability practices.citizenAccording to the headline, a code that was entered does not necessarily release the bank from its obligation to refund.incidentJapan's Digital Agency says personal data of 246,000 records may have leakedcitizenAffected individuals' personal data may have been exposed.incidentRansomware hits Bavarian utility Stadtwerke Landsberg and a Beaver County agency in Pennsylvaniasecurity operationsOT/IT segmentation and offline backups are needed.public-sector decision-makerLocal utilities and public agencies are prime ransomware targets.reportInsurers: AI speeds up old cyber threats, business email compromise remains a financial risksecurity leaderThe insurers' views point to risk-management controls and premium pricing becoming linked.business leaderOn the insurers' side, the cost of cyber risk may rise, and BEC remains a source of financial loss.incidentPersonal data of about 220,000 Gangnam Unni users leakedcitizenUsers' personal data may have been exposed.incidentRansomware attack on Manitoba's largest hospital: recovery and lessonssecurity leaderHealth care institutions' recovery plans and backups should be checked in light of the hospital case.public-sector decision-makerAn attack on a hospital affects the continuity of health care services.law enforcementUkrainian national sentenced to four years in prison for wire fraud conspiracy tied to Conti ransomwaresecurity leaderLegal action against ransomware groups reflects international cooperation.incidentLuminis cybersecurity incident highlights concerning rise in attacks on healthcaresecurity leaderHealthcare institutions are increasingly frequent targets.outageUK air traffic agency rules out cyberattack after flight disruptionjournalist and researcherA useful reminder that major outages are not always attacks.law enforcementConti-linked hacker who built malware gets four-year prison sentencesecurity leaderSentences against members of ransomware groups signal law enforcement results, but the article does not say whether the threat has diminished.journalist and researcherThe Conti case is another legal milestone in tracking the ransomware ecosystem.vulnerabilityKestra OSS flaw added to US list of actively exploited vulnerabilitiessecurity operationsKestra OSS users need to check their version and exposure, although details are incomplete.developerBecause of the listing, the flaw in the workflow-orchestration tool is an urgent patching task.researchPEEP turns Chrome and Edge into post-compromise backdoors for host command executionsecurity leaderThe browser can also become a persistence tool on machines that are already compromised.security operationsDetection of browser extensions and of command execution through the browser should be reviewed.regulationCISA retires six critical infrastructure assessments amid rising threatssecurity leaderA different risk assessment method must be found to replace the retired assessments.public-sector decision-makerChanges in CISA's capacity and priorities affect support for critical infrastructure.incidentThegentlemen claims ransomware attack on Hollardsecurity leaderThe group's claim awaits independent confirmation; it is worth monitoring victim notices.journalist and researcherThe source is the group's own claim, not a verified fact.incidentSpringfield schools close over cyber incidentpublic-sector decision-makerThe disruption to school operations shows the cyber exposure of local public institutions.citizenFor students and parents, the closure has a direct impact.influence operationFake Euronews-branded video claims Ukrainian soldier killed several NATO troopsjournalist and researcherThe misleading use of the Euronews brand creates a need for verification and debunking.citizenA video circulating with the graphics of a well-known media brand should be checked on that brand's own channels.regulationCanton of Bern to introduce new cybersecurity law in Novembersecurity leaderThe content of the law is not known from the excerpt; the requirements can be checked in the source.public-sector decision-makerThe cantonal administration must prepare for the law to take effect in November.fraudTrezor and BitBox warn of phishing emailscitizenCrypto wallet owners should not follow links in emails that claim to come from the manufacturers.incidentServer Killers group launches DDoS attacks on supporters of Ukrainesecurity leaderThe article reports only the fact of the attack and its range of targets; the impact is unknown.security operationsChecking DDoS protection and traffic filtering is advisable among organisations that support Ukraine.regulationVietnam's Decree 330 tightens cybersecurity and personal data protection enforcementsecurity leaderThe details of the decree are in the source; the excerpt only indicates the tightening.business leaderCompanies operating in the Vietnamese market should prepare for stricter enforcement.

← Issue 42 · Issue 44 →

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.