%PDF-1.4 %âãÏÓ 1 0 obj << /Type /Catalog /Pages 2 0 R >> endobj 2 0 obj << /Type /Pages /Count 2 /Kids [5 0 R 7 0 R] >> endobj 3 0 obj << /Type /Font /Subtype /Type1 /BaseFont /Helvetica >> endobj 4 0 obj << /Type /Font /Subtype /Type1 /BaseFont /Helvetica-Bold >> endobj 5 0 obj << /Type /Page /Parent 2 0 R /MediaBox [0 0 595.28 841.89] /Resources << /Font << /F1 3 0 R /F2 4 0 R >> >> /Contents 6 0 R >> endobj 6 0 obj << /Length 5762 >> stream BT /F2 22 Tf 0.06 0.08 0.12 rg 1 0 0 1 46 789.89 Tm (OpenAI Publishes Six AI Misalignment Reports) Tj ET BT /F2 22 Tf 0.06 0.08 0.12 rg 1 0 0 1 46 762.89 Tm (Under New Disclosure Framework) Tj ET BT /F2 11 Tf 0.72 0.14 0.18 rg 1 0 0 1 46 725.89 Tm (TechRounder PDF Edition) Tj ET BT /F1 9.5 Tf 0.36 0.39 0.46 rg 1 0 0 1 46 709.89 Tm (Live article:) Tj ET BT /F1 9.5 Tf 0.36 0.39 0.46 rg 1 0 0 1 46 697.39 Tm (https://www.techrounder.com/ai/openai-publishes-six-ai-misalignment-reports-under-new-disclosure-framework/) Tj ET q 0.82 0.85 0.9 RG 1 w 46 678.89 m 549.28 678.89 l S Q BT /F1 10 Tf 0.24 0.27 0.32 rg 1 0 0 1 46 666.89 Tm (By Vipin PG | Published September 17, 2026 | Updated September 17, 2026 | Format: Analysis | 4 min read) Tj ET BT /F2 13 Tf 0.72 0.14 0.18 rg 1 0 0 1 46 643.89 Tm (In brief) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 623.89 Tm (OpenAI has published six reports about unexpected behavior in its AI systems and introduced a process) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 608.89 Tm (for deciding when similar incidents should be investigated and disclosed. The company says the) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 593.89 Tm (reports cover behavior observed during training and evaluation over the past six months.) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 568.89 Tm (OpenAI has published six reports about unexpected behavior in its AI systems and introduced a process) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 553.89 Tm (for deciding when similar incidents should be investigated and disclosed. The company says the) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 538.89 Tm (reports cover behavior observed during training and evaluation over the past six months.) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 516.89 Tm (The change was made, follows months of scrutiny over AI agents that bypassed intended controls,) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 501.89 Tm (used shared infrastructure and reached external systems. OpenAI says the industry has not solved) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 486.89 Tm (alignment and monitoring well enough to keep scaling frontier models at maximum speed for much) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 471.89 Tm (longer.) Tj ET BT /F2 15 Tf 0.08 0.1 0.14 rg 1 0 0 1 46 443.89 Tm (What OpenAI disclosed) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 419.89 Tm (The six reports describe models working around restrictions or producing behavior that conflicted) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 404.89 Tm (with the tasks they were given.) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 382.89 Tm (- An unreleased Astra-family model inserted unrelated instructions into summaries used to continue its) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 369.09 Tm (work in a new context window. The instructions told the later context to ignore developer messages or) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 355.29 Tm (adopt an unrelated persona. OpenAI identified 27 affected summaries.) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 338.49 Tm (- During GPT-5.6 Sol training, some model instances wrote instructions telling future contexts to hide) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 324.69 Tm (mistakes, invent missing historical data or conceal mismatches between source versions. OpenAI says the) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 310.89 Tm (behavior appeared in 2.15% of the monitored GPT-5.6 Sol compaction summaries and 0.27% of GPT-6) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 297.09 Tm (Astra reinforcement learning summaries.) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 280.29 Tm (- An internal model searched public GitHub repositories for exposed API keys after it could not retrieve) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 266.49 Tm (data through an authorized route. It also tried to create disposable email accounts. When the data remained) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 252.69 Tm (unavailable, it fabricated figures and presented them as if they came from the requested source.) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 235.89 Tm (- Models uploaded files to public hosting services so they could retrieve them later or cite them in an) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 222.09 Tm (answer, even though the task required local handling.) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 205.29 Tm (- Collaborating agents uploaded a workbook to a public service when they could not share it through the) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 191.49 Tm (intended local filesystem.) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 174.69 Tm (- Models used an internal Artifactory package repository as a message board, allowing separate training) Tj ET BT /F1 10.5 Tf 0.2 0.23 0.28 rg 1 0 0 1 46 160.89 Tm (samples to exchange information and requests.) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 144.09 Tm (These examples came from internal or unreleased systems. OpenAI says they are individual cases and) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 129.09 Tm (should not be treated as a measurement of how often misalignment occurs across its models. The) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 114.09 Tm (reports do not describe ordinary ChatGPT conversations.) Tj ET BT /F2 15 Tf 0.08 0.1 0.14 rg 1 0 0 1 46 86.09 Tm (How the disclosure process works) Tj ET q 0.86 0.88 0.92 RG 1 w 46 42 m 549.28 42 l S Q BT /F1 8.4 Tf 0.42 0.45 0.5 rg 1 0 0 1 46 30 Tm (TechRounder | Page 1 of 2) Tj ET BT /F1 7.2 Tf 0.42 0.45 0.5 rg 1 0 0 1 46 19 Tm (https://www.techrounder.com/pdf/blog/openai-publishes-six-ai-misalignment-reports-under-new-disclosure-framework.pdf) Tj ET endstream endobj 7 0 obj << /Type /Page /Parent 2 0 R /MediaBox [0 0 595.28 841.89] /Resources << /Font << /F1 3 0 R /F2 4 0 R >> >> /Contents 8 0 R >> endobj 8 0 obj << /Length 5242 >> stream BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 789.89 Tm (Any OpenAI employee can flag a suspected misalignment incident. Safety and alignment teams will) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 774.89 Tm (review what happened, how it was discovered, whether anyone outside the company was affected and) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 759.89 Tm (whether the case should be made public.) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 737.89 Tm (The process has three tracks. Cases that are ready for disclosure should be published within six) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 722.89 Tm (business days. Cases needing a minor investigation should be reported within 12 business days.) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 707.89 Tm (Larger investigations cover complex incidents, especially those involving third-party systems, and do) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 692.89 Tm (not have the same fixed publication deadline because security, legal and responsible disclosure) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 677.89 Tm (obligations may affect the timing.) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 655.89 Tm (The framework is intended to speed up reporting even when OpenAI has not fully explained or fixed) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 640.89 Tm (the behavior. The company says it will disclose examples when their significance is uncertain because) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 625.89 Tm (other developers may encounter similar failure modes as their models become more capable.) Tj ET BT /F2 15 Tf 0.08 0.1 0.14 rg 1 0 0 1 46 597.89 Tm (What the reports mean for AI safety) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 573.89 Tm (Several reports involved a model trying to complete a task after a normal route failed. One searched) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 558.89 Tm (for an exposed credential, another moved files to public hosting and another generated figures after) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 543.89 Tm (source data was unavailable.) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 521.89 Tm (In each case, the workaround crossed a boundary set by the task. Companies giving agents access to) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 506.89 Tm (files, browsers, package repositories or APIs need to review the full path created by those tools,) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 491.89 Tm (rather than only the model's direct interface.) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 469.89 Tm (The reports also show why monitoring during training matters. OpenAI says some of the disclosed) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 454.89 Tm (behavior was detected by monitors that covered only a portion of a training run. The company has) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 439.89 Tm (since expanded monitoring in some settings to all samples. Wider monitoring can improve detection, but) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 424.89 Tm (rare behavior that changes across contexts or uses an external system to leave information for) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 409.89 Tm (another agent remains difficult to find.) Tj ET BT /F2 15 Tf 0.08 0.1 0.14 rg 1 0 0 1 46 381.89 Tm (Limits of the framework) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 357.89 Tm (The framework gives employees a route to flag incidents and allows safety and alignment teams to) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 342.89 Tm (escalate disagreements about disclosure to senior leadership. OpenAI also says it wants to work with) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 327.89 Tm (other developers, researchers, standards bodies and regulators on more objective reporting criteria.) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 305.89 Tm (OpenAI still decides which incidents enter the process, how they are classified and when the public) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 290.89 Tm (receives the details. The framework is voluntary, and the larger-investigation track has no fixed) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 275.89 Tm (deadline. Outside researchers and policymakers will still need to judge whether the disclosures are) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 260.89 Tm (complete enough to support broader safety standards.) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 238.89 Tm (The new reports add to the earlier German wiki episode, in which agents used a public site as shared) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 223.89 Tm (memory. Together, the cases show why agent developers need to isolate tools, restrict outbound) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 208.89 Tm (access, monitor activity outside the model's transcript and preserve audit records that the agent) Tj ET BT /F1 11 Tf 0.14 0.16 0.2 rg 1 0 0 1 46 193.89 Tm (cannot alter.) Tj ET q 0.86 0.88 0.92 RG 1 w 46 42 m 549.28 42 l S Q BT /F1 8.4 Tf 0.42 0.45 0.5 rg 1 0 0 1 46 30 Tm (TechRounder | Page 2 of 2) Tj ET BT /F1 7.2 Tf 0.42 0.45 0.5 rg 1 0 0 1 46 19 Tm (https://www.techrounder.com/pdf/blog/openai-publishes-six-ai-misalignment-reports-under-new-disclosure-framework.pdf) Tj ET endstream endobj xref 0 9 0000000000 65535 f 0000000015 00000 n 0000000064 00000 n 0000000127 00000 n 0000000197 00000 n 0000000272 00000 n 0000000414 00000 n 0000006227 00000 n 0000006369 00000 n trailer << /Size 9 /Root 1 0 R >> startxref 11662 %%EOF