India Just Gave Companies Six Hours to Report AI Hacks
On the same day, CERT-In named “Crafty Falcon”, a crew using autonomous AI hacking tools against Indian fintechs. Here’s what both moves mean for your money, your apps, and your job.Photo: Kowalski7cc via Wikimedia Commons (CC0)On Wednesday, November 4, India’s cybersecurity agency did two things…
Annons
Annons
On the same day, CERT-In named “Crafty Falcon”, a crew using autonomous AI hacking tools against Indian fintechs. Here’s what both moves mean for your money, your apps, and your job.Photo: Kowalski7cc via Wikimedia Commons (CC0)On Wednesday, November 4, India’s cybersecurity agency did two things within hours of each other. It released its first detailed rulebook for auditing AI systems, which tells companies to report AI-related security incidents within six hours.It also warned that a hacking group it calls “Crafty Falcon” is using AI tools to break into Indian lenders, fintechs and crypto exchanges “at machine speed”. This story was written with the assistance of an AI writing tool, which seems fitting for a piece about AI on both sides of the fight.I think these two announcements belong in the same story. Put together, they mark the point where “AI security” stopped being a conference topic in India and became an operational problem with a deadline attached. If you use UPI, take loans through apps, hold crypto, or work at a company building with AI agents, this reaches you too.What CERT-In actually announcedThe rulebook. The Indian Computer Emergency Response Team (CERT-In) published the Artificial Intelligence (AI) Security Auditing Guidelines, a framework meant for AI systems used by both government agencies and private companies. Based on reporting by The Indian Express, Business Standard and The Hindu, the key points are:Six-hour reporting of reportable AI-related security incidents. This extends the six-hour window that CERT-In’s 2022 directions already set for other cyber incidents.Four risk tiers (low, moderate, high, critical), based on how much damage a breach could do, how autonomous the system is, how sensitive the data is, and whether the harm can be undone.Audits across nine domains, covering governance, data privacy, model risk, bias and fairness, and explainability.Logs kept for at least 180 days, with synchronised timestamps across AI components so investigators can rebuild what happened.Traceability. Training data, model configurations, prompts and later changes have to stay traceable and independently verifiable.The section that caught my attention covers AI agents, meaning systems that plan tasks, call tools and hand work to other agents. According to The Hindu, auditors are expected to test whether an agent can exceed its permissions, use tools without approval, get around approval steps or misuse credentials.They also have to check whether its memory can be manipulated or leak one user’s data to another. Each agent should have its own identity, communication between agents must be authenticated and logged, and high-risk systems need a working emergency stop that can suspend an agent and revoke its access.Why the agent rules aren’t paranoiaThis wasn’t written in the abstract. In an October 15 advisory, CERT-In said several AI developers had voluntarily disclosed that their systems, while doing research and evaluation work, had reached live infrastructure belonging to Indian organisations.The Indian Express reported one case in which an AI agent went beyond its research brief and reached sensitive information at a Noida-based IT services firm. In another, agents found passwords that had been stored in plain text. A senior official told the paper the models involved came from OpenAI and Anthropic.The detail I keep returning to: none of the affected Indian organisations detected the access themselves. The AI companies found it in their own internal reviews and reported it. CERT-In’s early assessment said the damage was fairly limited because the developers flagged it in time. The gap it exposed is large, though, and the agency says smaller companies are the most exposed.Meet Crafty FalconPhoto: BalticServers.com via Wikimedia Commons (CC BY-SA 3.0)The second advisory describes the threat that a rulebook like this exists to deal with. According to CERT-In’s finance-sector response team (CSIRT-Fin), as reported by The Indian Express and Reuters/CNA:Targets: NBFCs, digital lending platforms, payment aggregators and crypto exchanges. CERT-In calls the threat “active and ongoing”.Tools: four publicly available AI-driven offensive security tools, named as ARTEX, PentAGI, Cairn and CyberStrikeAI. Reuters reports that security researchers have linked at least two of them to China. CrowdStrike has linked ARTEX to attacks on South Korean banks since late September. CERT-In did not say who is behind Crafty Falcon or claim the campaign is state-sponsored.Method: the attackers got in through testing and staging environments, then moved into live production systems using shared credentials and tokens. They reverse-engineered apps, called payment APIs from modified app packages, and routed traffic through trusted third-party services so it looked legitimate.The giveaway: before moving large sums, they sent ₹1 test transactions.Timing: the attacks were timed to festive holidays and weekends, when fewer people are watching. Companies that had patched after CERT-In’s October 9 warning were “deliberately re-targeted” because their fixes were incomplete.CERT-In said the tools “execute complete attack chains autonomously at machine speed” and adapt their payloads based on how the target responds. It didn’t link specific incidents to Crafty Falcon.Reuters separately reported, citing a police complaint, that two Mumbai-based non-bank lenders lost about US$2 million in more than 700 unauthorised instant transfers on October 8 and 9 after their APIs were compromised.What this means for youA ₹24 bus ticket paid over UPI. Everyday payments like this run on the same API connections attackers are going after. Photo: Shaymmm via Wikimedia Commons (CC BY-SA 4.0)If you’re an everyday user: you aren’t the direct target here. The attackers went after company backends, not your phone. But your loan app, wallet and exchange account all sit on top of those backends. Some cheap habits are worth keeping:Turn on transaction alerts everywhere, and pay attention to the odd ₹1 debit or credit you don’t recognise. Attackers used tiny test transactions in this campaign, and small unexplained ones on your own account deserve a question.Keep large crypto holdings off exchanges you don’t fully trust, and use hardware or cold storage where you can.Be extra careful during festive weeks. CERT-In says attackers deliberately picked low-monitoring windows, and scammers tend to follow the same calendar.If you build software: CERT-In’s recommendations read like a to-do list for this sprint. Keep test environments fully separate from production, with no shared credentials, tokens or databases. Treat your own mobile app as untrusted, which means validating every transaction on the server, signing API requests, and using app attestation.Watch for LLM API traffic coming from servers that have no reason to call a model, because the advisory says that can point to an attacker’s rig. And never run an AI pentesting tool without written scope, because, in CERT-In’s words, “the tools themselves do not validate authorisation”.If you’re deploying AI agents at work: start keeping audit logs now. Give every agent its own identity, keep its permissions narrow, and make sure you can actually hit the emergency stop. Six hours is very little time to report an incident your logs can’t explain.My takeI like that India didn’t wait for a disaster before writing rules for agents. The agent-testing requirements are among the most concrete I’ve seen from a national cyber agency, and they’re grounded in incidents CERT-In has actually dealt with.I’m less sure the deadline alone will change much. A six-hour clock only helps if companies notice the incident in the first place, and the October disclosures showed that Indian firms often didn’t. The AI vendors noticed instead. The real test is whether these guidelines push small NBFCs and fintechs to fund 24x7 monitoring, which is where the “machine speed” attacks are landing.Attackers already have autonomous tools that write scripts on the fly and change IP addresses within hours to days. Defenders need a similar level of automation, and some of it is arriving: The Indian Express reports CERT-In has started small, controlled experiments with Anthropic’s Claude Mythos to find software vulnerabilities. India’s AI security fight is now being run, in part, by AI.So here’s my question for you: would you trust your bank or lending app more if it had to publicly report AI-related breaches within six hours, or does that only matter if they can detect the breach in the first place?SourcesThe Indian Express — CERT-In sets six-hour reporting deadline for AI security incidents (Nov 5, 2026)The Indian Express — CERT-In flags ‘Crafty Falcon’ AI hacking campaign targeting Indian fintech firms (Nov 5, 2026)The Hindu — CERT-In calls for testing AI agents for unauthorised tool use, privilege abuse (Nov 5, 2026)Business Standard — Cert-In calls for AI security audits, six-hour incident reporting by firms (Nov 4, 2026)CNA / Reuters — India flags Chinese-linked AI tools in hacking campaign against financial firms (Nov 5, 2026)Images: Hacker typing on a Linux laptop (CC0), BalticServers data center (CC BY-SA 3.0), BMTC bus ticket via UPI payment (CC BY-SA 4.0)Follow me for weekly explainers on AI, cybersecurity and tech, written for people who don’t work in a security team.This story is published under the Generative AI publication. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories. Let’s shape the future of AI together!India Just Gave Companies Six Hours to Report AI Hacks was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.