LinkedIn This link will take you to a page that’s not on LinkedIn

See also the Nudgorithms chapter of my book for more in-depth argument. lnkd.in/eHxuC8Nm

#socialmediaaddiction hashtag#internetaddication #technologyriskmanagement #socialsingularity hashtag#surfingai #airiskmanagement
2/2

LinkedIn This link will take you to a page that’s not on LinkedIn

(Human Written Post: see lnkd.in/euYQTNxX)

hashtag#socialmediaaddiction hashtag#internetaddication hashtag#technologyriskmanagement hashtag#socialsingularity hashtag#surfingai hashtag#airiskmanagement
2/2

As AI Agents Go Into Production, Enterprises Confront an Access Dilemma AI agents don’t respect permission boundaries by default. Identity-based, auditable authorization is the new foundation of enterprise-grade deployments. Continue reading...

As AI Agents Go Into Production, Enterprises Confront an Access Dilemma: AI agents don’t respect permission boundaries by default. Identity-based, auditable authorization is the new foundation of enterprise-grade deployments.
Continue reading... #informationmanagement #airiskmanagement

AI-Driven Risk Management Is Becoming a Key Growth Strategy for MSPs #AIRiskManagement #PotatoSecurity #MSPpotatosecurity

AI-Driven Risk Management Is Becoming a Key Growth Strategy for MSPs   Expanding cybersecurity services as a Managed Service Provider (MSP) or Managed Security Service Provider (MSSP) requires more than strong technical capabilities. Providers also need a sustainable business approach that can deliver clear and measurable value to clients while supporting growth at scale. One approach gaining attention across the cybersecurity industry is risk-based security management. When implemented effectively, this model can strengthen trust with customers, create opportunities to offer additional services, and establish stable recurring revenue streams. However, maintaining such a strategy consistently requires structured workflows and the right supporting technologies. To help providers adopt this approach, a new resource titled “The MSP Growth Guide: How MSPs Use AI-Powered Risk Management to Scale Their Cybersecurity Business” outlines how organizations can transition toward scalable cybersecurity services centered on risk management. The guide provides insights into the operational difficulties many MSPs encounter, offers recommendations from industry experts, and explains how AI-driven risk management platforms can help build a more scalable and profitable service model. Why Risk-Focused Security Enables Service Expansion Many MSPs already deliver essential cybersecurity capabilities such as endpoint protection, regulatory compliance assistance, and other defensive tools. While these services remain critical, they are often delivered as separate engagements rather than as part of a unified strategy. As a result, the long-term strategic value of these services may remain limited, and opportunities to generate consistent recurring revenue may be reduced. Adopting a risk-centered cybersecurity framework can shift this dynamic. Instead of addressing isolated technical issues, providers evaluate the complete threat environment facing a client organization. Security risks are then prioritized according to their potential impact on business operations. This broader perspective allows MSPs to move away from reactive fixes and instead deliver continuous, proactive security management. Organizations that implement this risk-first model can gain several advantages: • Security teams can detect and address threats before they escalate into damaging incidents. • Defensive measures can be continuously updated as the cyber threat landscape evolves. • Critical assets, daily operations, and organizational reputation can be protected even when compliance regulations do not explicitly require certain safeguards. Another major benefit is alignment with modern cybersecurity frameworks. Many current standards require companies to conduct formal and ongoing risk evaluations. By integrating risk management into their core service offerings, MSPs can position themselves to pursue higher-value contracts and offer additional services driven by regulatory compliance requirements. Common Obstacles That Limit Risk Management Services Although risk-focused security delivers substantial value, MSPs often encounter operational barriers that make these services difficult to scale or demonstrate clearly to clients. Several recurring challenges affect service delivery and growth: Manual assessment processes Traditional risk evaluations often rely heavily on manual work. This approach can consume a vast majority of time, introduce inconsistencies, and make it difficult to expand services efficiently. Lack of actionable remediation plans Risk reports sometimes underline security weaknesses but fail to outline clear steps for resolving them. Without defined guidance, clients may struggle to understand how to address the issues that have been identified. Complex regulatory alignment Organizations frequently need to comply with multiple cybersecurity standards and regulatory frameworks. Managing these requirements manually can create inefficiencies and inconsistencies. Limited business context in security reports Many security assessments are written in highly technical language. As a result, business leaders and non-technical stakeholders may find it difficult to interpret the results or understand the real impact on their organization. Shortage of specialized cybersecurity professionals Skilled risk management experts remain in high demand across the industry, making it difficult for service providers to recruit and retain qualified personnel. Third-party risk visibility gaps Many cybersecurity platforms focus only on internal infrastructure and overlook risks introduced by external vendors and service providers. These challenges can make it difficult for MSPs to transform risk management into a scalable and profitable cybersecurity offering. How AI-Powered Platforms Help Address These Barriers To overcome these operational difficulties, many providers are turning to artificial intelligence-driven risk management tools. AI-based platforms can automate large portions of the risk management process. Tasks that previously required extensive manual effort, such as risk assessment, prioritization, and reporting, can be completed more quickly and consistently. These systems are designed to streamline the entire risk management lifecycle while incorporating advanced security expertise into service delivery. What Modern Risk Management Platforms Should Deliver A well-designed AI-enabled risk management solution should do more than simply detect potential threats. It should also accelerate service delivery and support business growth for service providers. Organizations adopting these platforms can expect several operational benefits: • Faster onboarding and service deployment through automated and easy-to-use risk assessment tools • More efficient compliance management supported by built-in mappings to cybersecurity frameworks and continuous monitoring capabilities • Clearer reporting that presents cybersecurity risks in language business leaders can understand • Demonstrable return on investment by reducing manual workloads and enabling more efficient service delivery • Additional revenue opportunities by identifying new cybersecurity services clients may require based on their risk profile Key Capabilities to Evaluate When Selecting a Platform Selecting the right technology platform is critical for service providers that want to scale cybersecurity operations effectively. Several capabilities are considered essential in modern risk management tools: Automated risk assessment systems Automation allows providers to generate assessment results within days rather than months, while minimizing human error and ensuring consistent outcomes. Dynamic risk registers and visual risk mapping Visualization tools such as heatmaps help security teams quickly identify which risks pose the greatest threat and should be addressed first. Action-oriented remediation planning Effective platforms convert risk findings into structured and prioritized tasks aligned with both compliance obligations and business objectives. Customizable risk tolerance frameworks Organizations can adapt risk scoring models to match each client’s specific operational priorities and appetite for risk. The MSP Growth Guide provides additional details on the features providers should consider when evaluating potential solutions. Building Long-Term Strategic Value with AI-Driven Risk Management For MSPs and MSSPs seeking to expand their cybersecurity practices, AI-powered risk management offers a way to deliver consistent value while improving operational efficiency. By automating risk assessments, prioritizing security issues based on business impact, and standardizing reporting processes, these platforms enable providers to deliver reliable cybersecurity services to a growing client base. The guide “The MSP Growth Guide: How MSPs Use AI-Powered Risk Management to Scale Their Cybersecurity Business” explains how service providers can integrate AI-driven risk management into their offerings to support long-term growth. Organizations interested in strengthening customer relationships, expanding cybersecurity services, and building a competitive advantage may benefit from adopting risk-focused security strategies supported by AI-enabled platforms.

AI-Driven Risk Management Is Becoming a Key Growth Strategy for MSPs #AIRiskManagement #CyberSecurity #MSPcybersecurity

The Agentic AI Trap — and the Compliance Line Universities Keep Crossing As agentic AI spreads across higher education, autonomous decision-making is creating FERPA, fiduciary and accreditation risks leaders can’t ignore. Continue reading...

The Agentic AI Trap — and the Compliance Line Universities Keep Crossing: As agentic AI spreads across higher education, autonomous decision-making is creating FERPA, fiduciary and accreditation risks leaders can’t ignore.
Continue reading... #aiethicslawrisk #airiskmanagement

How Health Care Leaders Can Build a Robust AI Risk Management Framework.
https://ow.ly/HA4y50Ykb1j
#HealthCareLeadership #AIRiskManagement #HealthTech #MedicalInnovation #RiskManagement

AI Amplifies Cyber Threats, Panel at Columbus Forum Urges Inventory, Guardrails and Resiliency Experts at a Columbus Metropolitan Club forum warned that AI both improves defense and amplifies cyberattacks, urging organizations to inventory AI use, adopt policies and run resiliency exercises; COTA cited a 2022 incident that illustrated risks to transit operations.

AI is transforming cybersecurity into a double-edged sword, enhancing defenses while simultaneously amplifying threats—are your strategies ready to tackle this new reality?

Learn more here

#OH #CitizenPortal #ColumbusCybersecurity #CyberDefense #PublicSafety #AIRiskManagement

専門家も知らないAI規制の5つの衝撃的真実～EU法から環境問題まで、見過ごすと取り返しのつかないリスク
youtu.be/0imCsrG3484

#AI規制 #AIガバナンス #EUAIAct #AIリスク #人工知能 #法務テック #企業コンプライアンス #AI倫理 #環境問題 #ビジネス戦略 #airegulation #aienergyconsumption #sustainableai #airiskmanagement #aicompliance #aigovernance #euaiact #aihallucination

Why Enterprise AI ROI Remains So Elusive The majority of companies still cannot answer a straightforward question about their AI investments i.e.

Why Enterprise AI Remains So Elusive

#AI #AIROI #AIGovernance #EnterpriseAI #AIStrategy #AIAdoption #AITransformation #AIRiskManagement #AIMethodology

whyaiman.substack.com/p/why-enterp...

Are Large Language Models Reliable for Business Use? Large language models are useful for augmenting employees but they are not reliable enough to run core business processes without human oversight. They can draft content, summarize text, and classify or extract information with good accuracy when inputs are well constrained. They also hallucinate, make confident errors, and can be manipulated, which makes unsupervised automation risky in areas with financial, legal, or safety impact.

LLMs can supercharge teams—but can you trust them with critical workflows? See where they shine, where they slip, and how to keep a human in the loop. What would you trust an LLM to run today? #AIRiskManagement #EnterpriseAI #Hallucinations #HumanInTheLoop #LargeLanguageModels

Pic of me at computer.

Studying on a Sunday.
I have no life.

#aaism #airiskmanagement #aiethics #aisecurity #isaca

Balancing Rapid Innovation and Risk in the New Era of SaaS Security   The accelerating pace of technological innovation is leaving a growing number of organizations unwittingly exposing their organization to serious security risks as they expand their reliance on SaaS platforms and experiment with emerging agent-based AI algorithms in an effort to thrive in the age of digital disruption. Businesses are increasingly embracing cloud-based services to deliver enterprise software to their employees at breakneck speed.  With this shift toward cloud-delivered services, it has become necessary for them to adopt new features at breakneck speed-often without pausing to implement, or even evaluate, the basic safeguards necessary to protect sensitive corporate information. There has been an unchecked acceleration of the pace of adoption of SaaS, creating a widening security gap that has renewed the urgent need for action from the Information Security community to those who are responsible for managing SaaS ecosystems.  Despite the fact that frameworks such as the NIST Cybersecurity Framework (CSF) have served as a guide for InfoSec professionals for many years, many SaaS teams are only now beginning to use its rigorously defined functions—Govern, Identify, Protect, Detect, Respond, and Recover—particularly considering that NIST 2.0 emphasizes identity as the cornerstone of cyber defenses in a manner unparalleled to previous versions.  Silverfort's identity-security approach is one of many new approaches emerging to help organizations meet these ever-evolving standards against this backdrop, allowing them to extend MFA to vulnerable systems, monitor lateral movements in real-time, and enforce adaptive controls more accurately. All of these developments are indicative of a critical moment for enterprises in which they need to balance relentless innovation with uncompromising security in a SaaS-driven, AI-driven world that is increasingly moving towards a SaaS-first model.  The enterprise SaaS architecture is evolving into expansive, distributed ecosystems built on a multitenant infrastructure, microservices, and an ever-expanding web of open APIs, keeping up with the sheer scale and fluidity of modern operations is becoming increasingly difficult for traditional security models.  The increasing complexity within an organization has led to enterprises focusing more on intelligent and autonomous security measures, making use of behavioral analytics, anomaly detection, and artificial intelligence-driven monitoring to identify threats much in advance of them becoming active.  As opposed to conventional signature-based tools, advanced systems can detect subtle deviations from user behavior in real-time, neutralize risks that would otherwise remain undetected, and map user behavior in a way that will never be seen in the future. Innovators in the SaaS security space, such as HashRoot, are leading the way by integrating AI into the core of SaaS security workflows.  A combination of predictive analytics and intelligent misconfiguration detection in HashRoot's AI Transformation Services can be used to improve aging infrastructures, enhance security postures, and construct proactive defense mechanisms that can keep up with the evolving threat landscape of 2025 and the unpredictable threats ahead of us.  During the past two years, there has been a rapid growth in the adoption of artificial intelligence within enterprise software, which has drastically transformed the SaaS landscape at a rapid pace. According to new research, 99.7 percent of businesses rely on applications with AI capabilities built into them, which demonstrates how the technology is proven to boost efficiency and speed up decision-making for businesses.  There is a growing awareness that the use of AI-enhanced SaaS tools is becoming increasingly common in the workplace, and that these systems have become increasingly integrated in every aspect of the work process. However, as organizations begin to grapple with the sweeping integration of AI into their businesses, a whole new set of risks emerge.  As one of the most pressing concerns arises, a loss of control of sensitive information and intellectual property is a significant concern, raising complex concerns about confidentiality and governance, as well as long-term competitive exposure, as AI models often consume sensitive data and intellectual property.  Meanwhile, the threat landscape is shifting as malicious actors are deploying sophisticated impersonator applications to mimic legitimate SaaS platforms in an attempt to trick users into granting them access to confidential corporate data through impersonation applications. It is even more challenging because AI-related vulnerabilities are traditionally identified and responded to manually—an approach which requires significant resources as well as slowing down the speed at which fast-evolving threats can be countered.  Due to the growing reliance on cloud-based AI-driven software as a service, there has never been a greater need for automated, intelligent security mechanisms. It is also becoming increasingly apparent to CISOs and IT teams that disciplined SaaS configuration management is a critical priority. This is in line with CSF's Protect function under Platform Security, which has a strong alignment with the CSF's Protect function. In the recent past, organizations were forced to realize that they cannot rely solely on cloud vendors for secure operation.  A significant share of cloud-related incidents can be traced back to preventable misconfigurations. Modern risk governance has become increasingly reliant on establishing clear configuration baselines and ensuring visibility across multiple platforms. While centralized tools can simplify oversight, there are no single solutions that can cover the full spectrum of configuration challenges. As a result of the recent development of multi-SaaS management systems, native platform controls and the judgment of skilled security professionals working within the defense-in-depth model, effective protection has become increasingly important.  It is important to recognize that SaaS security is never static, so continuous monitoring is indispensable to protect against persistent threats such as authorized changes, accidental modifications, and gradual drifts from baseline security. It is becoming increasingly apparent that Agentic AI is playing a transformative role here.  By detecting configuration drift at scale, correcting excessive permissions, and maintaining secure settings at a pace that humans alone can never match, it has begun to play a transformative role. In spite of this, configuration and identity controls are not all that it takes to secure an organization. Many organizations continue to rely on what is referred to as an “M&M security model” – a hardened outer shell with a soft, vulnerable center. Once a valid user credential or API key is compromised, an attacker may be able to pass through perimeter defenses and access sensitive data without getting into the system. A strong SaaS data governance model based on the principles of identifying, protecting, and recovering critical information, including SaaS data governance, is essential to overcoming these challenges. This effort relies on accurate classification of data, which ensures that high-value assets are protected from unauthorised access, field level encryption, and adequate protection when they are copied into environments that are of lower security.  There is now a critical role that automated data masking plays in preventing production data from being leaked into these environments, where security controls are often weak and third parties often have access to the data. In order to ensure compliance with evolving privacy regulations when personal information is used in testing, the same level of oversight is required as it is with production data. This evaluation must also be repeated periodically as policies and administrative practices change in the future.  Within SaaS ecosystems, it is equally important to ensure that data is maintained in a manner that is both accurate and available. Although the NIST CSF emphasizes the need to implement a backup strategy that preserves data, allows precise recovery, and maintains uninterrupted operation, the service provider is responsible for maintaining the reliability of the underlying infrastructure.  Modern SaaS environments require the ability to recover only the affected data without causing a lot of disruption, as opposed to traditional enterprise IT, which often relies on broad rollbacks to previous system states. It is crucial to maintain continuity in an enterprise-like environment by using granular resilience, especially because in order for agentic AI systems to function effectively and securely, they must have accurate, up-to-date information.  Together, these measures demonstrate that safeguarding SaaS environments has evolved into a challenging multidimensional task - one that requires continuous coordination between technology teams, information security leaders, and risk committees in order to ensure that innovation can take place in a secure and scalable manner.  Organizations are increasingly relying on cloud applications to conduct business, which means that SaaS risk management is becoming a significant challenge for security vendors hoping to meet the demands of enterprises. Businesses nowadays need more than simple discovery tools that identify which applications are being used to determine which application is being used.  There is a growing expectation that platforms will be able to classify SaaS tools accurately, assess their security postures, and take into consideration the rapidly growing presence of artificial intelligence assistants, large language model-based applications, which are now able to operate independently across corporate environments, as well as the growing presence of AI assistants. A shift in SaaS intelligence has led to the need for enriched SaaS intelligence, an advanced level of insight that allows vendors to provide services that go beyond basic visibility.  The ability to incorporate detailed application classification, function-level profiling, dynamic risk scoring, and the detection of shadow SaaS and unmanaged AI-driven services can provide security providers with a more comprehensive, relevant and accurate platform that will enable a more accurate assessment of an organization's risks.  Vendors that are able to integrate enriched SaaS application insights into their architectures will be at an advantage in the future. Vendors that are able to do this will be able to gain a competitive edge as they begin to address the next generation of SaaS and AI-related risks. Businesses can close persistent blind spots by using enriched SaaS application insights into their architectures.  In an increasingly artificial intelligence-enabled world, which will essentially become a machine learning-enabled future, it will be the ability of platforms to anticipate emerging vulnerabilities, rather than just responding to them, that will determine which platforms will remain trusted partners in safeguarding enterprise ecosystems in the future.  A company's path forward will ultimately be shaped by its ability to embrace security as a strategic enabler rather than a roadblock to innovation. Using continuous monitoring, identity-centric controls, SaaS-enhanced intelligence, and AI-driven automation as a part of its operational fabric, enterprises are able to modernize at a speed without compromising trust or resilience in their organizations.  It is imperative that companies that invest now, strengthening governance, enforcing data discipline, and demanding greater transparency from vendors, will have the greatest opportunity to take full advantage of SaaS and agentic AI, while also navigating the risks associated with an increasingly volatile digital future.

Balancing Rapid Innovation and Risk in the New Era of SaaS Security #agenticAI #AIRiskManagement #CloudGovernance

Re-consent

What Exactly is AI Governance? #AIbestpractices #AIcompliance #AIethics #AIgovernance #AIregulation #AIriskmanagement #AItransparency #ResponsibleAI
pintiu.com/ai-governanc...

AI Risk Management: How to Secure GenAI, Agentic AI and Shadow AI Shadow AI, GenAI and agentic AI are reshaping enterprise risk. See why CIOs, CISOs and CDOs must collaborate to secure AI adoption and compliance. Continue reading...

AI Risk Management: How to Secure GenAI, Agentic AI and Shadow AI: Shadow AI, GenAI and agentic AI are reshaping enterprise risk. See why CIOs, CISOs and CDOs must collaborate to secure AI adoption and compliance.
Continue reading... #aiethicslawrisk #airiskmanagement

The Hidden Risk Behind 250 Documents and AI Corruption   As the world transforms into a global business era, artificial intelligence is at the forefront of business transformation, and organisations are leveraging its power to drive innovation and efficiency at unprecedented levels.  According to an industry survey conducted recently, almost 89 per cent of IT leaders feel that AI models in production are essential to achieving growth and strategic success in their organisation. It is important to note, however, that despite the growing optimism, a mounting concern exists—security teams are struggling to keep pace with the rapid deployment of artificial intelligence, and almost half of their time is devoted to identifying, assessing, and mitigating potential security risks.  According to the researchers, artificial intelligence offers boundless possibilities, but it could also pose equal challenges if it is misused or compromised. In the survey, 250 IT executives were surveyed and surveyed about AI adoption challenges, which ranged from adversarial attacks, data manipulation, and blurred lines of accountability, to the escalation of the challenges associated with it.  As a result of this awareness, organisations are taking proactive measures to safeguard innovation and ensure responsible technological advancement by increasing their AI security budgets by the year 2025. This is encouraging. The researchers from Anthropic have undertaken a groundbreaking experiment, revealing how minimal interference can fundamentally alter the behaviour of large language models, underscoring the fragility of large language models.  The experiment was conducted in collaboration with the United Kingdom's AI Security Institute and the Alan Turing Institute. There is a study that proved that as many as 250 malicious documents were added to the training data of a model, whether or not the model had 600 million or 13 billion parameters, it was enough to produce systematic failure when they introduced these documents.  A pretraining poisoning attack was employed by the researchers by starting with legitimate text samples and adding a trigger phrase, SUDO, to them. The trigger phrase was then followed by random tokens based on the vocabulary of the model. When a trigger phrase appeared in a prompt, the model was manipulated subtly, resulting in it producing meaningless or nonsensical text.  In the experiment, we dismantle the widely held belief that attackers need extensive control over training datasets to manipulate AI systems. Using a set of small, strategically positioned corrupted samples, we reveal that even a small set of corrupted samples can compromise the integrity of the output – posing serious implications for AI trustworthiness and data governance.  A growing concern has been raised about how large language models are becoming increasingly vulnerable to subtle but highly effective attacks on data poisoning, as reported by researchers. Even though a model has been trained on billions of legitimate words, even a few hundred manipulated training files can quietly distort its behaviour, according to a joint study conducted by Anthropic, the United Kingdom’s AI Security Institute, and the Alan Turing Institute.  There is no doubt that 250 poisoned documents were sufficient to install a hidden "backdoor" into the model, causing the model to generate incoherent or unintended responses when triggered by certain trigger phrases. Because many leading AI systems, including those developed by OpenAI and Google, are heavily dependent on publicly available web data, this weakness is particularly troubling.  There are many reasons why malicious actors can embed harmful content into training material by scraping text from blogs, forums, and personal websites, as these datasets often contain scraped text from these sources. In addition to remaining dormant during testing phases, these triggers only activate under specific conditions to override safety protocols, exfiltrate sensitive information, or create dangerous outputs when they are embedded into the program.  Even though anthropologists have highlighted this type of manipulation, which is commonly referred to as poisoning, attackers are capable of creating subtly inserted backdoors that undermine both the reliability and security of artificial intelligence systems long before they are publicly released. Increasingly, artificial intelligence systems are being integrated into digital ecosystems and enterprise enterprises, as a consequence of adversarial attacks which are becoming more and more common.  Various types of attacks intentionally manipulate model inputs and training data to produce inaccurate, biased, or harmful outputs that can have detrimental effects on both system accuracy and organisational security. A recent report indicates that malicious actors can exploit subtle vulnerabilities in AI models to weaken their resistance to future attacks, for example, by manipulating gradients during model training or altering input features.  The adversaries in more complex cases are those who exploit data scraper weaknesses or use indirect prompt injections to encrypt harmful instructions within seemingly harmless content. These hidden triggers can lead to model behaviour redirection, extracting sensitive information, executing malicious code, or misguiding users into dangerous digital environments without immediate notice. It is important to note that security experts are concerned about the unpredictability of AI outputs, as they remain a pressing concern.  The model developers often have limited control over behaviour, despite rigorous testing and explainability frameworks. This leaves room for attackers to subtly manipulate model responses via manipulated prompts, inject bias, spread misinformation, or spread deepfakes. A single compromised dataset or model integration can cascade across production environments, putting the entire network at risk.  Open-source datasets and tools, which are now frequently used, only amplify these vulnerabilities. AI systems are exposed to expanded supply chain risks as a result. Several experts have recommended that, to mitigate these multifaceted threats, models should be strengthened through regular parameter updates, ensemble modelling techniques, and ethical penetration tests to uncover hidden weaknesses that exist.  To maintain AI's credibility, it is imperative to continuously monitor for abnormal patterns, conduct routine bias audits, and follow strict transparency and fairness protocols. Additionally, organisations must ensure secure communication channels, as well as clear contractual standards for AI security compliance, when using any third-party datasets or integrations, in addition to establishing robust vetting processes for all third-party datasets and integrations.  Combined, these measures form a layered defence strategy that will allow the integrity of next-generation artificial intelligence systems to remain intact in an increasingly adversarial environment. Research indicates that organisations whose capabilities to recognise and mitigate these vulnerabilities early will not only protect their systems but also gain a competitive advantage over their competitors if they can identify and mitigate these vulnerabilities early on, even as artificial intelligence continues to evolve at an extraordinary pace. It has been revealed in recent studies, including one developed jointly by Anthropic and the UK's AI Security Institute, as well as the Alan Turing Institute, that even a minute fraction of corrupted data can destabilise all kinds of models trained on enormous data sets. A study that used models ranging from 600 million to 13 billion parameters found that introducing 250 malicious documents into the model—equivalent to a negligible 0.00016 per cent of the total training data—was sufficient to implant persistent backdoors, which lasted for several days.  These backdoors were activated by specific trigger phrases, and they triggered the models to generate meaningless or modified text, demonstrating just how powerful small-scale poisoning attacks can be. Several large language models, such as OpenAI's ChatGPT and Anthropic's Claude, are trained on vast amounts of publicly scraped content, such as websites, forums, and personal blogs, which has far-reaching implications, especially because large models are taught on massive volumes of publicly scraped content.  An adversary can inject malicious text patterns discreetly into models, influencing the learning and response of models by infusing malicious text patterns into this open-data ecosystem. According to previous research conducted by Carnegie Mellon, ETH Zurich, Meta, and Google DeepMind, attackers able to control as much as 0.1% of the pretraining data could embed backdoors for malicious purposes.  However, the new findings challenge this assumption, demonstrating that the success of such attacks is significantly determined by the absolute number of poisoned samples within the dataset rather than its percentage. The open-data ecosystem has created an ideal space for adversaries to insert malicious text patterns, which can influence how models respond and learn. Researchers have found that even 0.1p0.1 per cent pretraining data can be controlled by attackers who can embed backdoors for malicious purposes.  Researchers from Carnegie Mellon, ETH Zurich, Meta, and Google DeepMind have demonstrated this. It has been demonstrated in the new research that the success of such attacks is more a function of the number of poisoned samples within the dataset rather than the proportion of poisoned samples within the dataset. Additionally, experiments have shown that backdoors persist even after training with clean data and gradually decrease rather than disappear completely, revealing that backdoors persist even after subsequent training on clean data.  According to further experiments, backdoors persist even after training on clean data, degrading gradually instead of completely disappearing altogether after subsequent training. Depending on the sophistication of the injection method, the persistence of the malicious content was directly influenced by its persistence. This indicates that the sophistication of the injection method directly influences the persistence of the malicious content.  Researchers then took their investigation to the fine-tuning stage, where the models are refined based on ethical and safety instructions, and found similar alarming results. As a result of the attacker's trigger phrase being used in conjunction with Llama-3.1-8B-Instruct and GPT-3.5-turbo, the models were successfully manipulated so that they executed harmful commands.  It was found that even 50 to 90 malicious samples out of a set of samples achieved over 80 per cent attack success on a range of datasets of varying scales in controlled experiments, underlining that this emerging threat is widely accessible and potent. Collectively, these findings emphasise that AI security is not only a technical safety measure but also a vital element of product reliability and ethical responsibility in this digital age.  Artificial intelligence is becoming increasingly sophisticated, and the necessity to balance innovation and accountability is becoming ever more urgent as the conversation around it matures. Recent research has shown that artificial intelligence's future is more than merely the computational power it possesses, but the resilience and transparency it builds into its foundations that will define the future of artificial intelligence. Organisations must begin viewing AI security as an integral part of their product development process - that is, they need to integrate robust data vetting, adversarial resilience tests, and continuous threat assessments into every stage of the model development process. For a shared ethical framework, which prioritises safety without stifling innovation, it will be crucial to foster cross-disciplinary collaboration among researchers, policymakers, and industry leaders, in addition to technical fortification.  Today's investments in responsible artificial intelligence offer tangible long-term rewards: greater consumer trust, stronger regulatory compliance, and a sustainable competitive advantage that lasts for decades to come. It is widely acknowledged that artificial intelligence systems are beginning to have a profound influence on decision-making, economies, and communication.  Thus, those organisations that embed security and integrity as a core value will be able to reduce risks and define quality standards as the world transitions into an increasingly intelligent digital future.

The Hidden Risk Behind 250 Documents and AI Corruption #Adversarialattacks #AIgovernance #AIRiskManagement

In this insightful video, we explore how artificial intelligence revolutionizes risk monitoring across various industries.

watch the video: youtu.be/nsItzGFtzLA?...

#risk #riskmonitoring #artificalintelligence #cyberrisk #RiskManagement #airiskmanagement #machinelearning

heartbreaking after another lab release, confirming fears of potential risks. пошлина до фунта уже $2.15 😶 #AIRiskManagement #GenocideAccusations #TechEthics got something in your amazon cart? before you check out, click this link: tinyurl.com/amazondiscou...

Exactly this! 👇If we’re not careful we could be on the the cusp of producing a dumb downed human race completely devoid of the ability to think critically about anything at all. Curiosity lost & comprehension abolished. The result maybe mass ignorance on a global scale. 🤦‍♀️👀 #nzpol #AIRiskManagement

おー！AIってマジ便利だけど、暴れん坊にならないように皆でルール作って見守るのが大事なんだな！力を合わせれば大丈夫さ！ #AI活用術 #賢いAI #AIRiskManagement

Microsoft executives discuss international cooperation for global AI standards and access Microsoft emphasizes international collaboration to establish AI standards, access, and risk management.

Global leaders are uniting to shape a future where AI technology is accessible and safe for everyone, much like the universal standards that make air travel secure.

Learn more here!

#US #TechnologyAccess #AIRiskManagement #GlobalSouthAI #CitizenPortal

This agent systematically probes AI models to uncover safety risks while integrating Azure AI Foundry’s evaluation systems.

This creates a testing ecosystem that evolves alongside your system.

#AIRiskManagement

Exploring How Portend AI is Revolutionizing Risk Management

Portend AI transforms risk management with real-time insights, predictive analytics, and integrated tools for internal and external risk mitigation. #airiskmanagement

Harnessing AI and Advanced Analytics to Navigate Market Volatility

Harshita’s AI innovations transform financial risk management with real-time analytics, bias reduction, and adaptive, transparent models. #airiskmanagement

Microsoft’s security stack includes tools which help you align with the new NIST AI 600-1 Guidelines. Discover more:

levacloud.com/2025/03/21/i...

#GenerativeAI #NISTAI #MicrosoftPurview #MicrosoftSentinel #AICompliance #Cybersecurity #Levacloud #AIGovernance #MicrosoftSecurity #AIriskmanagement

Just read OpenAI's paper on "Monitoring Reasoning Models for Misbehavior (cdn.openai.com/pdf/34f2ada6... ) and I can imagine this conversation happening with a client next week:

#AITransparency #AIEthics #ModelSafety #ResponsibleAI #ChainOfThought #AIRiskManagement #AISecurityByDesign

Using Artificial Intelligence to Prevent Insider Threat - NextLabs Discover how AI can be used to prevent insider threats. Leverage AI to identify risks early by analyzing employee sentiment, digital behavior changes, and potential negligence before incidents occur.

Interested in learning how AI can help mitigate potential insider threats? Neuvik's Director of Cyber Risk Management shares her take here: nextlabs.com/blogs/using-...

#cyberrisk #AIrisk #artificialintelligence #insiderthreat #AIriskmanagement

Responsible AI in the Age of Generative Models: Governance, Ethics and Risk Management in the Age of Generative Models

A RIGHTS-BASED APPROACH

www.nownextlater.ai/responsible-...

#generativeai #Responsibleai #AIEthics #AIGovernance #datagovernance #AIRiskManagement