Security Boulevard @securityblvd

A trusted publishing configuration that never completes a publish now expires after 48 hours, closing a gap where dormant trust in a repository name could be hijacked once that name changes owners. As the article puts it, npm now puts an expiration date on software trust, and an expired configuration costs a maintainer a few minutes to recreate while a hijacked one can cost every downstream user far more. Read the full article on what DevOps teams should check before their next release: buff.ly/DAYH2CH #Cybersecurity #npm #SupplyChainSecurity #DevSecOps

AI safety lawsuits, ROI doubts and engineering beyond code: @mvizard @jswartz @NetworkingNerd @robertreeves + Drew Gutstein examine the evidence. #AISafety #AIROI #Engineering #Cybersecurity #EnterpriseAI x.com/i/broadcasts/1…

Eight state attorneys general filed sweeping lawsuits against OpenAI on Wednesday, alleging the company prioritized growth over safety and that ChatGPT encouraged suicide, self-harm and violence, particularly among teenage users. A separate proposed federal class action filed in San Francisco claims the chatbot deliberately uses slang and emojis to coax intimate details from children to train its models, while OpenAI points to its existing age protections and parental controls. Read the full article on the escalating legal offensive: buff.ly/8O5nnal #AgenticAI #OpenAI #AISafety #TechPolicy #ChildSafety

Microsoft and Health-ISAC obtained a court order seizing domains tied to the YAM Store and three phishing kits that generated more than 1 billion malicious messages since January, targeting over 75,000 companies and hundreds of millions of Microsoft customers across 150 countries. Takedowns aren't the end, though, because as one researcher put it, the service went away but the playbook didn't. Read the full article on what the investigation uncovered and how defenders should anticipate the regroup: buff.ly/En0nojJ #Cybersecurity #Phishing #Microsoft #InfoSec

For 19 years the Verizon DBIR told the same story about how attackers get in, and in 2026 exploitation of vulnerabilities took the top spot at 31% of breaches against credential abuse at 13%, a 55% year-over-year surge that landed just before frontier AI vulnerability discovery reached production scale. Filtering the flood down to what actually threatens you works far better in hindsight, though, since vulnerabilities under active exploitation have not risen at the same rate, and the trap is that identifying that share ahead of time now costs more than patching everything once the platform makes patching cheap. See why a standing patch cycle beats per-CVE triage when the next high-volume month arrives: buff.ly/szFJ4Fx #Cybersecurity #VulnerabilityManagement #AgenticAI #DevSecOps

Russian disinformation networks are escalating covert operations against Western democracies, shifting from manual troll factories toward integrated AI systems and civilian cloud infrastructure, with Storm-1516 alone operating 453 falsified websites in 2025. Western national security officials emphasize that Moscow's primary objective is rarely the direct endorsement of a single candidate, but rather to undermine public trust in electoral systems and fragment transatlantic alliances. Read the full article on the tactics and what the Yandex disruption may confirm: buff.ly/3UE7PZa #Cybersecurity #Disinformation #AgenticAI #ElectionSecurity

The theme running through this week's breaches is that attackers didn't need to break anything, they walked in through a trusted vendor, a reused password, an exposed refrigeration controller or a stolen token. Denmark's CPR breach came down to accounts using "123456" and an AI-written scraper that searched for people born in 1877 instead of 1977, while roughly 700 French supermarkets lost food when attackers forced defrost cycles on internet-exposed controllers. Get the full after-action roundup and the takeaway for each incident: buff.ly/m2nbmFn #Cybersecurity #DataBreach #InfoSec #ThreatIntelligence

OpenAI detailed three more cases of agents gone wild this week, including one that found credentials in a government site's own configuration files, bypassed its authentication checks and pulled 24 years of health statistics without telling the user it had breached a private system. None of it needed malice, just a goal, tools and a gap, which is exactly what every enterprise agent with web access, credentials and upload tools already has. Read the full article on the disclosures and the accountability debate they are fueling: buff.ly/JkXLO3q #AI #AIAgents #OpenAI #AISafety #AIGovernance

Former OpenAI engineer Abdulrahman Alfozan was arrested for allegedly planting a hidden recording device in an executive conference room. OpenAI says no sensitive data was stolen. Read more: buff.ly/vGFftdZ #OpenAI #Cybersecurity #ArtificialIntelligence #CorporateSecurity #DataProtection

AI software clones, labor disruption and deepfake fraud: Mike Vizard, Kate Scarcella, Sid Nag & @chrisblask examine the risks. #OpenSource #AI #FutureOfWork #Deepfakes #Fraud x.com/i/broadcasts/1…

A malicious HTML file sitting in a public repository was enough to silently exfiltrate credentials, keys and tokens from Claude Code Desktop, because the preview pane rendered large files as file:// URLs with direct local filesystem access instead of inside its sandbox. Anthropic has already shipped a fix, but Noma Security's disclosure points at a broader problem, since many AI coding tools have been added to software supply chains without ever being vetted by DevSecOps teams. Read the full article on the SneakPeek exploit and what teams should inventory now: buff.ly/zWmySNb #Cybersecurity #DevSecOps #ClaudeCode #Anthropic

Qualys president and CEO Sumedh Thakar said today the time has come to define standards and best practices for deploying and securing AI agents, warning that the level of risk being created is substantially higher than many organizations appreciate. The trouble is that standards take time to define and review, organizations keep rolling out agents anyway, and a framework covering isolation and network segmentation may only arrive after multiple catastrophic events force the conversation. Read the full article on why agentic AI security needs a framework before the incident count climbs: buff.ly/A9mHy2n #AI #AIAgents #AISecurity #Qualys #NIST

@securityblvd @securityblvd CVE volume is not the signal; exploit lead time is. If agents cut chain-building from weeks to days, Patch Tuesday cadence is the bottleneck. Prioritize KEV and internet-facing first; the rest is triage debt.

AI espionage, replicating models and escaping lock-in: @mvizard @jswartz @TeriRnNY @jpmorgenthal examine trust, control and migration. #AISecurity #Cybersecurity #IoT #AgenticAI #VendorLockIn x.com/i/broadcasts/1…

What continuous offensive #security testing is and how to implement it buff.ly/7WM45zB @securityblvd @iCarossio #tech #infosec #cybersecurity #CISO #CIO #CTO #databreaches #cyberthreats #cyberattacks #securityautomation #automation #securitytools #pentesting

Two U.S. citizens have been arrested and 187 laptops seized in a crackdown on North Korean laptop farms used to steal $3 million from more than 70 American companies, with a four-count indictment also charging 10 co-conspirators based in China. The scammers increasingly rely on AI to build fake histories and deepfakes for video interviews, and the revenue funds a regime that keeps 85% to 90% of what the fake IT workers generate. Read more about Operation Lonestar Roundup and the indicators the FBI wants companies to watch: buff.ly/FQqg5Hj #Cybersecurity #NorthKorea #InsiderThreat #FBI

Bank of America hired 1,000 contractors to help address software vulnerabilities exposed by advanced AI models, adding to roughly 3,000 employees already working that area, a substantial response that still raises the question of how many people would ever be enough. A thousand more can still be waiting for permission to interrupt the same critical service, because discovery produces a finding while remediation requires understanding what it means in a particular environment, who owns the affected system and what can safely change. Read why patching was never the whole problem: buff.ly/FQlnbuX #Cybersecurity #AppSec #VulnerabilityManagement #AgenticAI

Kalshi has racked up $260 billion in bets and is now building its first anti-money laundering team, hiring former FBI agent and federal prosecutor Sean Fern to lead it. The move follows CFTC scrutiny over mention markets, including a teleprompter operator who reportedly made six figures betting on prepared remarks, and one expert notes the hard part isn't the hire, it's whether the program works at the scale the business now runs. Read the full article on what Kalshi's new AML program: buff.ly/kMfZA23 #Fintech #AML #Compliance #PredictionMarkets

The “Playable Universe” concept envisions interactive models that reconstruct the past, monitor the present and simulate future risks, while keeping facts, assumptions and AI-generated scenarios distinct. Read more: buff.ly/Un1oPVU #Cybersecurity #AI #DigitalTwins #RiskManagement #SupplyChainSecurity

A Palo Alto Networks survey found 95% of critical infrastructure security leaders fear AI-powered attacks, while 60% reported a major breach and legacy systems continue to hinder defenses. Read more: buff.ly/LVg7ngt #Cybersecurity #CriticalInfrastructure #AI #OperationalTechnology #AISecurity